PDA

View Full Version : LW 2015 vs LW 2018 Arch Viz render testing



Pages : [1] 2

Nicolas Jordan
01-24-2018, 10:48 AM
Here is a Arch Viz production scene that I'm sharing publicly for testing purposes. It takes almost twice as long to render in LW 2018 vs LW 2015 from my initial testing. Can this scene be rendered in 2018 at or very close to the same speed as in 2015 while still retaining the overall quality in areas such as AA, lighting, GI and reflection blurring?

Feel free to download this LW 2015 project and render it in 2018. If you are able to get it to render faster and close to LW 2015 times on your machine please reshare here so we can all see and learn how to optimize LW 2018 scenes.


Scene notes:

The light comes in as a distant light in 2018 and should be increased to about 3.14 to match the 2015 scene.

Objects such as grass, trees, shrubs and the house have local GI settings to control and optimize GI.

I have replaced all commercial assets with ones that are found for free online. The trees and shrubs are free from Xfrog and the sky is free from Viz People.

This scene requires DP Sunsky plugin which can be downloaded here http://dpont.pagesperso-orange.fr/plugins/Sunsky.html

Here is a public link to the scene from my dropbox https://www.dropbox.com/s/ucjwzv4g71etkab/House_Rendering_Test.zip?dl=0

Rayek
01-24-2018, 12:01 PM
Thank you for the test scene. I am rendering it now in 2015. When I opened the file, it couldn't find two Juniper tree textures (replaced with the alternate ones in the textures folder), and it requires the DP Sunsky plugin (which I had not installed).

Out of curiosity, what are your machine specs and respective render times?

Nicolas Jordan
01-24-2018, 12:53 PM
Thank you for the test scene. I am rendering it now in 2015. When I opened the file, it couldn't find two Juniper tree textures (replaced with the alternate ones in the textures folder), and it requires the DP Sunsky plugin (which I had not installed).

Out of curiosity, what are your machine specs and respective render times?

Oops sorry about the missing images. I just found them and copied the files so the I think everything should be there now. Yes I also forgot to mention the DP Sunsky plugin is required. I will update my notes on the original post.

I'm rendering on a i7 4930K. My render time in 2015 is 16m 4 sec. In 2018 it was just over 30 min.

Rayek
01-24-2018, 01:03 PM
2015 version rendered in 29 minutes. My i7 is getting old. See my sig for hardware.

jeric_synergy
01-24-2018, 01:14 PM
Oops sorry about the missing images.
I'm interested: did you use "Package Scene" to make that archive? Wondering if it missed those images.

Nicolas Jordan
01-24-2018, 01:43 PM
I'm interested: did you use "Package Scene" to make that archive? Wondering if it missed those images.

No I didn't use package scene because it usually crashes for me so I just did it manually. Package scene does work properly from time to time for me but I have found that it's not very stable.

jwiede
01-24-2018, 01:47 PM
Well, off to a less-than-fun start here: LW2018 converted the scene (minus Sunsky) into a LW2018 scene without obvious problem. However, the converted scene now 100%-reliably crashes LW2018 to desktop when F9-rendered (crash is always Render Worker thread getting a SIGSEGV at recurring offset in "MipmapFilteredPixel", precise thread # and accessed location varies). To be clear, this is just setting content dir, loading the converted scene, and hitting F9, nothing else.

Anyone else encountering crash with converted LW2018 version of scene just attempting an F9?

mummyman
01-24-2018, 02:14 PM
Bummer things are taking so much longer to render. Is there any rumblings of making LW work with GPU other than Octane? Has anyone taken this scene into Octane and do some render times?

Nicolas Jordan
01-24-2018, 02:31 PM
Well, off to a less-than-fun start here: LW2018 converted the scene (minus Sunsky) into a LW2018 scene without obvious problem. However, the converted scene now 100%-reliably crashes LW2018 to desktop when F9-rendered (crash is always Render Worker thread getting a SIGSEGV at recurring offset in "MipmapFilteredPixel", precise thread # and accessed location varies). To be clear, this is just setting content dir, loading the converted scene, and hitting F9, nothing else.

Anyone else encountering crash with converted LW2018 version of scene just attempting an F9?

I didn't experience any crashes when rendering the converted 2018 scene.

Rayek
01-24-2018, 02:47 PM
I stopped rendering in LW 2018 - it was taking too long. Will play with this later this week.

Nicolas Jordan
01-24-2018, 03:05 PM
Bummer things are taking so much longer to render. Is there any rumblings of making LW work with GPU other than Octane? Has anyone taken this scene into Octane and do some render times?

Here is a link to a similar exterior scene that was done in Octane and posted. http://forums.newtek.com/showthread.php?155498-Show-your-LW2018-Test-Renders&p=1533752&viewfull=1#post1533752

I doubt Octane would be much faster unless you had lots of video cards.

jboudreau
01-24-2018, 08:52 PM
Here is a Arch Viz production scene that I'm sharing publicly for testing purposes. It takes almost twice as long to render in LW 2018 vs LW 2015 from my initial testing. Can this scene be rendered in 2018 at or very close to the same speed as in 2015 while still retaining the overall quality in areas such as AA, lighting, GI and reflection blurring?

Feel free to download this LW 2015 project and render it in 2018. If you are able to get it to render faster and close to LW 2015 times on your machine please reshare here so we can all see and learn how to optimize LW 2018 scenes.


Scene notes:

The light comes in as a distant light in 2018 and should be increased to about 3.14 to match the 2015 scene.

Objects such as grass, trees, shrubs and the house have local GI settings to control and optimize GI.

I have replaced all commercial assets with ones that are found for free online. The trees and shrubs are free from Xfrog and the sky is free from Viz People.

This scene requires DP Sunsky plugin which can be downloaded here http://dpont.pagesperso-orange.fr/plugins/Sunsky.html

Here is a public link to the scene from my dropbox https://www.dropbox.com/s/ucjwzv4g71etkab/House_Rendering_Test.zip?dl=0

Hi,

Just downloaded your scene, It says it's missing this texture EU48brk1.tif

Thanks,
Jason

Nicolas Jordan
01-24-2018, 09:08 PM
Hi,

Just downloaded your scene, It says it's missing this texture EU48brk1.tif

Thanks,
Jason

I found it and updated the file so it should be there now. Hopefully there are no others missing.

Kryslin
01-25-2018, 12:55 AM
My first go with this scene had a respectable 28m 23s, on a 6 YO i7 (stats in sig, slower than Rayek's machine). Got a little bit of noise and some fireflies, so I'm starting to tweak things.
I'll need to work on it tomorrow and on the weekend.

Be interesting to see what RH comes up with...

rustythe1
01-25-2018, 02:52 AM
this took 3 mins on my old i7
139716
some of the materials still need work, but that will probably just make it faster as if you notice the ceader board is quite reflective at the moment (next to the entrance door)

rustythe1
01-25-2018, 03:28 AM
this one even quicker as i realised there was an illuminous sphere so changed for backdrop and sampled in the environment light
139717

samurai_x
01-25-2018, 07:47 AM
The scene looks cartoony. Especially the last two with very harsh shadows and no bounced light.

jboudreau
01-25-2018, 07:48 AM
this one even quicker as i realised there was an illuminous sphere so changed for backdrop and sampled in the environment light
139717

How many threads do you have, What version of lightwave are you using? Also how long did it take to render in 2015? compared to 2018. Also can you share your scene?

P.S. Something is wrong with the cedar planks on the house. The texture is not fitting properly. Also it looks like you have no ray traced reflection on the doors and windows because they are really dark

Thanks,
Jason

alexos
01-25-2018, 08:09 AM
So on my 5-yrs old dual Xeon (24 cores total) 2015 took about 12 minutes, 2018.01... 29 minutes. That is slow, even though the image admittedly looks better. I've changed lights to something more 2018-friendly (a distant sun-like and an environment to replace Sunsky) and didn't bother with the rest. I'm sure tweaking surfaces and suchlike would make it quicker, but I doubt it would reach 2015's times anyway.

ADP.

139720

tyrot
01-25-2018, 08:17 AM
you can make this render in 10 seconds with a 200 dollars second hand GTX 780 .. and it will look 100 times better.. if someone making archviz (especially exteriors) please do not hurt your business .. buy octane..

THIBAULT
01-25-2018, 08:29 AM
you can make this render in 10 seconds with a 200 dollars second hand GTX 780 .. and it will look 100 times better.. if someone making archviz (especially exteriors) please do not hurt your business .. buy octane..

Totaly agree ! Don't loose your time with lightwave 2018 render isn't archiviz friendly !

And soon, Octane have denoiser !

https://www.facebook.com/groups/OctaneRender/permalink/934581923385553/

Nicolas Jordan
01-25-2018, 09:08 AM
So on my 5-yrs old dual Xeon (24 cores total) 2015 took about 12 minutes, 2018.01... 29 minutes. That is slow, even though the image admittedly looks better. I've changed lights to something more 2018-friendly (a distant sun-like and an environment to replace Sunsky) and didn't bother with the rest. I'm sure tweaking surfaces and suchlike would make it quicker, but I doubt it would reach 2015's times anyway.

ADP.

139720

I spent lots of time playing around with surfaces in 2018 both with standard and BSDF but it really didn't seem to make much difference in my render times.

jboudreau
01-25-2018, 09:21 AM
So on my 5-yrs old dual Xeon (24 cores total) 2015 took about 12 minutes, 2018.01... 29 minutes. That is slow, even though the image admittedly looks better. I've changed lights to something more 2018-friendly (a distant sun-like and an environment to replace Sunsky) and didn't bother with the rest. I'm sure tweaking surfaces and suchlike would make it quicker, but I doubt it would reach 2015's times anyway.

ADP.

139720

Try turning off Mipmapping in your render globals, this makes a world of difference. For some reason mipmapping is causing a white outline around the leaves in the trees and in the grass. It also is causing a lot of noise and terrible AA. If you turn it off the picture is not nearly as noisy so it allows you to decrease your samples and AA therefore speeding up your renders. This will also make your trees and grass much nicer and closer to the original image.

Hope this helps

Thanks,
Jason

Nicolas Jordan
01-25-2018, 09:25 AM
this one even quicker as i realised there was an illuminous sphere so changed for backdrop and sampled in the environment light
139717

Something definitely doesn't look right overall. There is no GI or less GI not sure but something looks off. It would be interesting to know what your settings are.

jboudreau
01-25-2018, 09:37 AM
Here is what I was referring to above,

I rendered a close up of the tree leaves with mipmapping on and off

Here it is with Mipmapping = On

http://animatrixproductions.com/Trees_Mipmapping_On.png


Here it is with Mipmapping = Off

http://animatrixproductions.com/Trees_Mipmapping_Off.png

Thanks,
Jason

Nicolas Jordan
01-25-2018, 11:00 AM
Here is what I was referring to above,

I rendered a close up of the tree leaves with mipmapping on and off

Here it is with Mipmapping = On

http://animatrixproductions.com/Trees_Mipmapping_On.png


Here it is with Mipmapping = Off

http://animatrixproductions.com/Trees_Mipmapping_Off.png

Thanks,
Jason

Interesting discovery, I will have to give that a try as soon as I have a chance.

RebelHill
01-25-2018, 01:06 PM
So, honestly... Im not sure how good an example this scene is. Truth is its a LONG way from being optimal in 2015 Im guessing... at least for the quality. Its very flat lighting, theres little shading detail, I dont think ud have anything vastly different if youd binned of GI and just used ambient light with some occlusion. IMO its really overdone in may ways (in terms of rendertime costly stuff) with respect to the actual final appearance.

Anyhow, the 2015 version took me 29mins to render... my 2018 version here, took 14. So a little more than twice as fast. Is it different, sure... I didnt really bother to go through and do over all the materials, couple small changes to one or 2, most of what Ive done is just to alter the light and render setup. Generally speaking, its much the same "quality" as the original, and substantially faster... so there it is, once again... 2018 gives the same level of quality, or better, in the same time or less as compared to 2015.

Skonk
01-25-2018, 01:56 PM
So, honestly... Im not sure how good an example this scene is. Truth is its a LONG way from being optimal in 2015 Im guessing... at least for the quality. Its very flat lighting, theres little shading detail, I dont think ud have anything vastly different if youd binned of GI and just used ambient light with some occlusion. IMO its really overdone in may ways (in terms of rendertime costly stuff) with respect to the actual final appearance.

Anyhow, the 2015 version took me 29mins to render... my 2018 version here, took 14. So a little more than twice as fast. Is it different, sure... I didnt really bother to go through and do over all the materials, couple small changes to one or 2, most of what Ive done is just to alter the light and render setup. Generally speaking, its much the same "quality" as the original, and substantially faster... so there it is, once again... 2018 gives the same level of quality, or better, in the same time or less as compared to 2015.

Once you have 2 years of experience with the current version.

Nicolas Jordan
01-25-2018, 02:24 PM
So, honestly... Im not sure how good an example this scene is. Truth is its a LONG way from being optimal in 2015 Im guessing... at least for the quality. Its very flat lighting, theres little shading detail, I dont think ud have anything vastly different if youd binned of GI and just used ambient light with some occlusion. IMO its really overdone in may ways (in terms of rendertime costly stuff) with respect to the actual final appearance.

Anyhow, the 2015 version took me 29mins to render... my 2018 version here, took 14. So a little more than twice as fast. Is it different, sure... I didnt really bother to go through and do over all the materials, couple small changes to one or 2, most of what Ive done is just to alter the light and render setup. Generally speaking, its much the same "quality" as the original, and substantially faster... so there it is, once again... 2018 gives the same level of quality, or better, in the same time or less as compared to 2015.

I have to say it looks washed out. It looks like it's being lit by a pure white environment. IMO the overall quality is much worse than the original but this may be partially to do with the surfaces not being adjusted.

Also something I have noticed in my testing is render times in 2018 increase dramatically with blurred reflections or "glossy reflections as it is now called.

It is clear to me now that the 2018 render engine is vastly different under the hood and while it makes some things easier to accomplish it can make other very difficult and the render times are much more unpredictable than 2015. I have used other render engines in production before but 2018 is a vastly different beast to wrestle with.

RebelHill
01-25-2018, 02:25 PM
I have to say it looks washed out. It looks like it's being lit by a pure white environment. IMO the overall quality is much worse than the original but this may be partially to do with the surfaces not being adjusted.

Surfaces arent adjusted... could be I set the lighting stronger... the "wash out" is not a "render quality" thing... merely a "saturation" thing. Tweaking and adjusting that would not require going anywhere near sampling, GI, etc.
Its really the shading/lighting/etc quality you;re looking for from the renderer, not the specific colour of a given texture.

And here's something else with I think is pretty instructive of the basic principles (and gets us away from me having to spend ages trying to match the look of various materials). Strip away all textures and special surfacing whatevers... 50% grey plain surface, everything else the same... 2015 ugly as sin and 5m15s... 2018 so much nicer and only 1m48.

I rest my case.


Once you have 2 years of experience with the current version.

It didnt take that long to learn such things u know. I pretty much figured all this out in the space of a couple weeks through various systematic tests of the different options, and that was before many of the features were fully finished and before there was any documentation. Its not hard.

jboudreau
01-25-2018, 02:26 PM
So, honestly... Im not sure how good an example this scene is. Truth is its a LONG way from being optimal in 2015 Im guessing... at least for the quality. Its very flat lighting, theres little shading detail, I dont think ud have anything vastly different if youd binned of GI and just used ambient light with some occlusion. IMO its really overdone in may ways (in terms of rendertime costly stuff) with respect to the actual final appearance.

Anyhow, the 2015 version took me 29mins to render... my 2018 version here, took 14. So a little more than twice as fast. Is it different, sure... I didnt really bother to go through and do over all the materials, couple small changes to one or 2, most of what Ive done is just to alter the light and render setup. Generally speaking, its much the same "quality" as the original, and substantially faster... so there it is, once again... 2018 gives the same level of quality, or better, in the same time or less as compared to 2015.

Can you share your scene for download?

Thanks,
Jason

jboudreau
01-25-2018, 02:30 PM
So, honestly... Im not sure how good an example this scene is. Truth is its a LONG way from being optimal in 2015 Im guessing... at least for the quality. Its very flat lighting, theres little shading detail, I dont think ud have anything vastly different if youd binned of GI and just used ambient light with some occlusion. IMO its really overdone in may ways (in terms of rendertime costly stuff) with respect to the actual final appearance.

Anyhow, the 2015 version took me 29mins to render... my 2018 version here, took 14. So a little more than twice as fast. Is it different, sure... I didnt really bother to go through and do over all the materials, couple small changes to one or 2, most of what Ive done is just to alter the light and render setup. Generally speaking, its much the same "quality" as the original, and substantially faster... so there it is, once again... 2018 gives the same level of quality, or better, in the same time or less as compared to 2015.

Also I think you might have the wrong color space for that scene since the cedar wood on the house is washed out. You need to change your Picked colors to SRGB and Light Color to SRGB since all the images in his scene have been set to SRGB. At least that's what I had to do get no washout on the cedar wood surface

Thanks,
Jason

RebelHill
01-25-2018, 02:38 PM
Can you share your scene for download?

Done


Also I think you might have the wrong color space for that scene since the cedar wood on the house is washed out...

Yeah, maybe... point is its irrelevant. The sat/darkness/CS of textures has NO bearing on the rendertime, shading effects and so on, only the "colourcast".

jboudreau
01-25-2018, 02:41 PM
Done



Yeah, maybe... point is its irrelevant. The sat/darkness/CS of textures has NO bearing on the rendertime, shading effects and so on, only the "colourcast".

Yeah that's true, I was just pointing it out where they were saying the render looked washed out.

Can you let me know what the specs of your system is. Are you on 8 threads or higher?

Thanks,
Jason.

jwiede
01-25-2018, 02:42 PM
Anyhow, the 2015 version took me 29mins to render... my 2018 version here, took 14. So a little more than twice as fast. Is it different, sure... I didnt really bother to go through and do over all the materials, couple small changes to one or 2, most of what Ive done is just to alter the light and render setup. Generally speaking, its much the same "quality" as the original, and substantially faster... so there it is, once again... 2018 gives the same level of quality, or better, in the same time or less as compared to 2015.

What's your config (CPU, RAM, OS, etc)? Was that a "load and hit F9" render in 2015.3?

A render time of 29mins is surprisingly long for this scene in 2015.3, considering an i7 4930 rendered it in 16min 4sec, and my aging MacPro 5,1 (2x 2.4g 'Westmere' Xeon E5645, 32GB RAM) rendered it in 14min 23sec (result attached).

139740

RebelHill
01-25-2018, 02:43 PM
Was that a "load and hit F9" render in 2015.3?

No I went and changed a whole bunch of stuff willy nilly so the comparison would be utterly meaningless.

Skonk
01-25-2018, 02:48 PM
I say this with the utmost respect for you; but you have a much better brain for this kind of stuff than the average person. The fact that your training videos are so well loved is testament to this. If it was as easy for everyone else as it is for you then no one would need your training; but they do need it.

The average joe is clearly having issues getting good results out of this new renderer.

jboudreau
01-25-2018, 02:48 PM
Done



Yeah, maybe... point is its irrelevant. The sat/darkness/CS of textures has NO bearing on the rendertime, shading effects and so on, only the "colourcast".

Can you please do a package scene all the models are missing?

Thanks,
Jason

RebelHill
01-25-2018, 02:59 PM
I say this with the utmost respect for you; but you have a much better brain for this kind of stuff than the average person. The fact that your training videos are so well loved is testament to this. If it was as easy for everyone else as it is for you then no one would need your training; but they do need it.

Well I have perhaps more of a understanding about what's going on behind the scenes... but that's not really all that. Mainly, its just process of elimination... there only are so many light types, there are only so many places to turn one set of samples up, another down. There are only so many ways to achieve a particular lighting effect (example exterior environmental illumination, there's BG GI, the same interpolated, or the enviro light), and so on. If you go through them, and their combinations METHODICALLY, and dont just try all mixes of settings in a chaotic manner... the answers reveal themselves without you having to really know anything.


The average joe is clearly having issues getting good results out of this new renderer.

I can only point to the comment Ive pointed to before...

http://forums.newtek.com/showthread.php?155474-LW-2018-Comments-Opinions&p=1532266&viewfull=1#post1532266

This REALLY is ALL you need to know. Its actually that simple.


Can you please do a package scene all the models are missing?

Ugh... cant be bothered, they're huge... plus I didnt really change anything... just increased roughness on the leaves n grass surfaces, otherwise as per the original import.

jboudreau
01-25-2018, 03:19 PM
Well I have perhaps more of a understanding about what's going on behind the scenes... but that's not really all that. Mainly, its just process of elimination... there only are so many light types, there are only so many places to turn one set of samples up, another down. There are only so many ways to achieve a particular lighting effect (example exterior environmental illumination, there's BG GI, the same interpolated, or the enviro light), and so on. If you go through them, and their combinations METHODICALLY, and dont just try all mixes of settings in a chaotic manner... the answers reveal themselves without you having to really know anything.



I can only point to the comment Ive pointed to before...

http://forums.newtek.com/showthread.php?155474-LW-2018-Comments-Opinions&p=1532266&viewfull=1#post1532266

This REALLY is ALL you need to know. Its actually that simple.



Ugh... cant be bothered, they're huge... plus I didnt really change anything... just increased roughness on the leaves n grass surfaces, otherwise as per the original import.

Okay but did you use the Principled BSDF surface or the standard one?

Thanks,
Jason

Nicolas Jordan
01-25-2018, 03:19 PM
Surfaces arent adjusted... could be I set the lighting stronger... the "wash out" is not a "render quality" thing... merely a "saturation" thing. Tweaking and adjusting that would not require going anywhere near sampling, GI, etc.
Its really the shading/lighting/etc quality you;re looking for from the renderer, not the specific colour of a given texture.

And here's something else with I think is pretty instructive of the basic principles (and gets us away from me having to spend ages trying to match the look of various materials). Strip away all textures and special surfacing whatevers... 50% grey plain surface, everything else the same... 2015 ugly as sin and 5m15s... 2018 so much nicer and only 1m48.

I rest my case.



It didnt take that long to learn such things u know. I pretty much figured all this out in the space of a couple weeks through various systematic tests of the different options, and that was before many of the features were fully finished and before there was any documentation. Its not hard.

Um I'm not sure what settings you have for the 2015 image but that is not correct. This is what I get in 2015 with the original settings when the scene is grey. It very similar to what you get in 2018.

RebelHill
01-25-2018, 03:21 PM
standard... I did switch the grass to principled, but just to see what "look" difference it made rather than because it changed anything timewise... so yeah, when I previously said, increased rough... I guess I meant reduced gloss.

RebelHill
01-25-2018, 03:40 PM
Um I'm not sure what settings you have for the 2015 image but that is not correct.

Yep... I must've messed something up there. Tried it again, and I get the same image you get... and now it takes me 6m43 to render.

2018s 1m40 odd is still SIGNIFICANTLY faster, and my case remains rested.

jboudreau
01-25-2018, 04:22 PM
Yep... I must've messed something up there. Tried it again, and I get the same image you get... and now it takes me 6m43 to render.

2018s 1m40 odd is still SIGNIFICANTLY faster, and my case remains rested.

Are you guys, Using OD Tools to strip out all the textures from the scene, If not how do you guys do this in 2018 since most of the surfaces are linked to nodes so the color channel is grey'd out

Also what are you specs for your system. You still haven't told us. 8 threads, 12 threads, 32 64 ..... etc.

Thanks,
Jason

RebelHill
01-25-2018, 04:23 PM
You just make 1 surface plain grey manually... copy it and paste over all the others.

jboudreau
01-25-2018, 04:33 PM
So, honestly... Im not sure how good an example this scene is. Truth is its a LONG way from being optimal in 2015 Im guessing... at least for the quality. Its very flat lighting, theres little shading detail, I dont think ud have anything vastly different if youd binned of GI and just used ambient light with some occlusion. IMO its really overdone in may ways (in terms of rendertime costly stuff) with respect to the actual final appearance.

Anyhow, the 2015 version took me 29mins to render... my 2018 version here, took 14. So a little more than twice as fast. Is it different, sure... I didnt really bother to go through and do over all the materials, couple small changes to one or 2, most of what Ive done is just to alter the light and render setup. Generally speaking, its much the same "quality" as the original, and substantially faster... so there it is, once again... 2018 gives the same level of quality, or better, in the same time or less as compared to 2015.

Hey man thanks for the scene.

You realize you don't have GI on in your scene. This is why you are getting so fast render times. It takes almost 4-5min in itself just to process the GI. Try turning on GI and see what times you get

Thanks,
Jason

Nicolas Jordan
01-25-2018, 04:37 PM
Hey man thanks for the scene.

You realize you don't have GI on in your scene. This is why you are getting so fast render times. It takes almost 4-5min in itself just to process the GI. Try turning on GI and see what times you get

Thanks,
Jason

Also I'm not sure if he had any blurry reflections on any surfaces which also can change the render time. The orginal 2015 scene has blurry reflections on the black metal as well as the siding. THis is definitely not an apples to apples comparison. I don't think anything has been proven or accomplished one way or the other yet.

RebelHill
01-25-2018, 04:40 PM
You realize you don't have GI on in your scene.

Oddly enough, I realise everything about the scene.

Let me state the whole point AGAIN...

The workflows and practices in 2018 are not the same as previous versions. There are tools and options that were not available before. If you try to do everything the exact same way you did it before, dont be surprised if it doesnt work out the way you hope. Do what works best in 2018 and it outperforms earlier versions.

- - - Updated - - -


Also I'm not sure if he had any blurry reflections on any surfaces which also can change the render time. The orginal 2015 scene has lots of blurry reflections.

The part earlier where I mentioned having increased roughness/lowered gloss... doesnt sound like a blurry reflection thing?

Yes I did have them.

Yes, I know what what Im doing.

jboudreau
01-25-2018, 04:42 PM
Oddly enough, I realise everything about the scene.

Let me state the whole point AGAIN...

The workflows and practices in 2018 are not the same as previous versions. There are tools and options that were not available before. If you try to do everything the exact same way you did it before, dont be surprised if it doesnt work out the way you hope. Do what works best in 2018 and it outperforms earlier versions.

I understand that

So what you are saying is we don't need GI for this scene in 2018 but we did in 2015.3, that seems a bit odd

Thanks,
Jason

- - - Updated - - -


Oddly enough, I realise everything about the scene.

Let me state the whole point AGAIN...

The workflows and practices in 2018 are not the same as previous versions. There are tools and options that were not available before. If you try to do everything the exact same way you did it before, dont be surprised if it doesnt work out the way you hope. Do what works best in 2018 and it outperforms earlier versions.

- - - Updated - - -



The part earlier where I mentioned having increased roughness/lowered gloss... doesnt sound like a blurry reflection thing?

Yes I did have them.

Yes, I know what what Im doing.

I know you know what you are doing, To be honest your settings that you have in your scene were very close to what I had been playing with but I had GI on and that drove the render time through the roof because it takes so long to process.

Thanks,
Jason

RebelHill
01-25-2018, 04:45 PM
So what you are saying is we don't need GI for this scene in 2018 but we did in 2015.3

Well lets see... its easiest to refer to the grey/clay render isnt it... 2015 uses GI, 2018 doesnt, but oh look... pretty soft background illumination. What could that possibly mean??

Did you notice the comment I linked back to before regarding best practices for 2018? Did you notice the part that said "dont bother with BG GI"? Could that be a clue?

How many times do I have to say the EXACT same things?

jboudreau
01-25-2018, 05:07 PM
Well lets see... its easiest to refer to the grey/clay render isnt it... 2015 uses GI, 2018 doesnt, but oh look... pretty soft background illumination. What could that possibly mean??

Did you notice the comment I linked back to before regarding best practices for 2018? Did you notice the part that said "dont bother with BG GI"? Could that be a clue?

How many times do I have to say the EXACT same things?

Sorry no I did not see that comment you linked back too.

Here is your scene with the necessary tweaks to some materials and the Color Space.

http://animatrixproductions.com/2018_House_NOGI.png

This took me 3 Minutes 17 seconds on a 8 thread laptop

139748

139749

Thanks,
Jason

gar26lw
01-25-2018, 06:36 PM
Once you have 2 years of experience with the current version.

mmm, something that bugged the he’ll outta me was when beta users were getting at people for complaining there were no bugfixes or updates or roadmap or communication, all the while sitting there playing away.

kinda irritating. (none of that aimed at you rh)

samurai_x
01-25-2018, 06:50 PM
All the renders still don't look anything close to photoreal. Not a good benchmark for render speed and quality.
Looks like kray(even v2) is still better for photoreal stills.

gar26lw
01-25-2018, 06:53 PM
All the renders still don't look anything close to photoreal. Not a good benchmark for render speed and quality.
Looks like kray(even v2) is still better for photoreal stills.

maybe someone can do one in kray and octane ?

jboudreau
01-25-2018, 06:59 PM
All the renders still don't look anything close to photoreal. Not a good benchmark for render speed and quality.
Looks like kray(even v2) is still better for photoreal stills.

Well I don't think we are spending a ton of time to make the scene look photo realistic. We are trying to get the scene to look the same as the original but bring the render times down drastically. I'm honestly shocked by just using a distant light with the environment light together with GI = Off in 2018 You can get almost exactly the same GI as you get in 2015 with GI = ON especially for exterior scenes and it renders extremely fast withing Seconds compared to Several minutes.

I might take the scene into octane later and post my results

Thanks,
Jason

samurai_x
01-25-2018, 07:03 PM
Well I don't think we are spending a ton of time to make the scene look photo realistic. We are trying to get the scene to look the same as the original but bring the render times down drastically. I'm honestly shocked by just using a distant light with the environment light together with GI = Off in 2018 You can get almost exactly the same GI as you get in 2015 with GI = ON especially for exterior scenes and it renders extremely fast withing Seconds compared to Several minutes.

I might take the scene into octane later and post my results

Thanks,
Jason

I don't use gi with outdoor scenes unless the highest quality is required. Dpsunsky in lw 2015 is the fastest way to fake gi.



I think this scene rendered almost 10 years ago is avail for testing at evermotion.

https://i.imgur.com/9adhRD6.jpg

https://i.imgur.com/ZvD0Z2W.jpg

jboudreau
01-25-2018, 07:04 PM
I don't use gi with outdoor scenes unless the highest quality is required. Dpsunsky in lw 2015 is the fastest way to fake gi.

Yeah that's what this scene is using in 2015 DPsunsky

samurai_x
01-25-2018, 07:14 PM
Yeah that's what this scene is using in 2015 DPsunsky

That's kind of redundant if gi and dpsunsky rre in the scene.

jboudreau
01-25-2018, 07:27 PM
That's kind of redundant if gi and dpsunsky rre in the scene.

It's not my scene man, I'm just testing it. Never really used DPsunsky before. I have octane so no need too

Thanks,
Jason

samurai_x
01-25-2018, 08:11 PM
It's not my scene man, I'm just testing it. Never really used DPsunsky before. I have octane so no need too

Thanks,
Jason

Never? Its so useful to fake gi. Its like the old 3dmax skylight for outdoor scenes that renders uber fast.
In lw 2015 and below you can just use dpsunsky+dome/direct to fake gi. Works great on animations, too. No gi artifacts or splotches.

jboudreau
01-25-2018, 08:21 PM
Never? Its so useful to fake gi. Its like the old 3dmax skylight for outdoor scenes that renders uber fast.
In lw 2015 and below you can just use dpsunsky+dome/direct to fake gi. Works great on animations, too. No gi artifacts or splotches.

Well I shouldn't say never, I have played around with it. I haven't really looked back at the native render ever since I got Octane

Thanks,
Jason

MichaelT
01-25-2018, 08:39 PM
Just for interest, I rendered two images, One with GI & one without. Just to see how much of an impact GI have.
As it turns out... very little. I don't doubt the same effects could be achieved by adding a few lights in select places.

Disabling GI to render a scene, really have a very substantial impact.

139750
139751

rustythe1
01-26-2018, 01:21 AM
exactly all i did, all gi did was make the render take 10 times longer with minimal to no affect, and i would say that things like corners and over hangs look a lot more realistic with the gi off, see this area.
139756
and that's more or less all i did, other small things were put things like reflection and refraction bounces, samples down as for that view you wouldn't have anything above 1 or 2 needed,

Myagi
01-26-2018, 03:08 AM
I'm just curious, isn't the comparison a bit in the wrong place so to speak. If I'm understanding it right, the 2018 env lights (and various backdrop setting) are basically a built-in method for LW to auto-generate a light probe image on-the-fly. So for a fairer comparison, in terms of render operation performed, it would mean using an "Image World" with a light probe image (and backdrop-only GI) in an older LW?

OTOH it's kind-of a PITA to create light probes, so it's great that 2018 can composite one internally based on bg setttings and light objects that you easily can change settings of (and even animate). So the time savings are rather in scene setup than necessarily in rendering?

RebelHill
01-26-2018, 03:31 AM
I'm just curious, isn't the comparison a bit in the wrong place so to speak. If I'm understanding it right, the 2018 env lights (and various backdrop setting) are basically a built-in method for LW to auto-generate a light probe image on-the-fly. So for a fairer comparison, in terms of render operation performed, it would mean using an "Image World" with a light probe image (and backdrop-only GI) in an older LW?

No, because new things have new stuff. As Ive said many times, if you're trying to "pilot" 2018 using the exact same methods and workflows as 2015 or earlier, then you're not using it to its design capacity, it's not the same renderer ad you shouldnt be surprised if you have very different results.

If, however, you use 2018 in "the 2018 way", you can get renders that are as good or better (quality wise) in the same rendertime or less, than 2015 in pretty much every case.

If you have a 2 core computer, and you upgrade to an 8 core... it makes little sense say, "well, the 8 core is only faster because of more cores, its not a meaningful comparison unless you turn 6 of them off". You really have to compare things based each on their own individual potential.

UnCommonGrafx
01-26-2018, 03:37 AM
Surely both, at the end of the day?

Getting the job faster is what we ought to want out of the next iteration of a software.

RebelHill
01-26-2018, 03:38 AM
mmm, something that bugged the heíll outta me was when beta users were getting at people for complaining there were no bugfixes or updates or roadmap or communication, all the while sitting there playing away.

Yeah... I never joined any of those discussions, not once did I say to anyone... shut your crying pieholes, snowflakes... Nor did I ever proclaim... chin up, chaps, itll be along eventually.

And in defence of the others, I can promise you that there were some pretty heated and occasionally unkind words spoken by many in the beta along the lines of "when tf is this thing getting released??" at times. The frustrations were quite universally shared.

RebelHill
01-26-2018, 03:40 AM
Getting the job faster is what we ought to want out of the next iteration of a software.

Indeed, and doing that may entail having to use new, previously unavailable tools, and/or a different approach and workflow, which might seem not directly comparable, but the important comparison is the result, not the workflow by which you achieve it.

Myagi
01-26-2018, 03:51 AM
No, because new things have new stuff. As Ive said many times, if you're trying to "pilot" 2018 using the exact same methods and workflows as 2015 or earlier, then you're not using it to its design capacity, it's not the same renderer ad you shouldnt be surprised if you have very different results.

That's kind of the opposite of what I was suggesting. What I was suggesting was changing the settings in old LW to match how 2018 is rendering it *, to achieve a more comparable render operation. And also as I noted, that wouldn't take anything away from the fact that 2018 is a great workflow improvement in that regard, even if the render time differences become smaller.

(Edit: * in the ~1 minute rendering)

UnCommonGrafx
01-26-2018, 03:59 AM
I am finding the workflow full of clicks but quite enjoyable based on my results. They have been amazing, thus far.
Everything works, now. Smoothly and predictably. And the lighting... nice!
FiberFx can be PLAYED with, now!?
And GI comes in a different flavor of lights and more power with less - less gi rays yet smoother and faster, when needed, with portal and environment lights pulling most of the duty in newer scenes.

Eh... to the choir...

RebelHill
01-26-2018, 04:09 AM
That's kind of the opposite of what I was suggesting. What I was suggesting was changing the settings in old LW to match how 2018 is rendering it, to achieve a more comparable render operation. And also as I noted, that wouldn't take anything away from the fact that 2018 is a great workflow improvement in that regard, even if the render time differences become smaller.

Ah ok... sure, I guess you could try and "backtrack" the workflow, that'd seem a fair approach. Difficult in the sense that some of the new things in 2018 dont really have comparitors available in 2015, but you could try for a best fit. And sure, rendertime differences may well become smaller. As I initially said on entering this thread, I think the 2015 scene is not very well put together to begin with... my 2018 in half the time is more extreme than I think should be the case. On scenes like this, optimised well in 2015 and the same in 2018... Id generally expect the latter to be 10-20% faster.

The phrase I keep on using that sums up my experience... the same quality or better, in the same time or less. There will be situations where 2018 is loads faster, some where it's only a little, or roughly the same, but there are almost no situations where 2018, used right, takes longer to render a given scene than previous versions.

rustythe1
01-26-2018, 04:09 AM
That's kind of the opposite of what I was suggesting. What I was suggesting was changing the settings in old LW to match how 2018 is rendering it *, to achieve a more comparable render operation. And also as I noted, that wouldn't take anything away from the fact that 2018 is a great workflow improvement in that regard, even if the render time differences become smaller.

(Edit: * in the ~1 minute rendering)

the old scene did have an HDRI environment in the form of a physical image sphere if you look back i noted it was there and replaced it with the environment light attached to the same image so the results should have been the same.

lightman
01-26-2018, 07:08 AM
Ok I'm late to the party, but I have been playing around with the renderer on my on models and scenes.
Here is the render from Nicolas
2015 default on my system is 15 minutes 10 seconds

I imported to 2018 changed some of the glossiness setting down to 5 percent or so and reduced the spec on some surfaces down to 4 percent.
Turned off GI and interpolated, unchecked all the options in GI, upped the reflection samples to 4 and refraction to 4.
On the direct light I set it to 3.14 on intensity. Added an enviro light and set it's samples to 8 and intensity to 1, sample backdrop to on and hit render.

LW2018 8 minutes 14 seconds
It looks very close to the 2015 version just needs some more adjustment on the spec of the materials, but the contact shadows are a lot better. Bump needs some adjustment to. I think over all its much better just takes a different way of thinking about the lighting compared to 2015.
139757139758

THIBAULT
01-26-2018, 08:01 AM
My test ! Core I7 6950 X - 20 Threads - 40 mn with LW 2018 - 1920 X 1080
HDRI - No GI
Happy but working with Octane in the studio and 40 mn in this format it's long............. Max 10 mn with Octane

139759

samurai_x
01-26-2018, 08:32 AM
My test ! Core I7 6950 X - 20 Threads - 40 mn with LW 2018 - 1920 X 1080
HDRI - No GI
Happy but working with Octane in the studio and 40 mn in this format it's long............. Max 10 mn with Octane

139759

That's a better test scene. Post it for lw 2105 and 2018

Nicolas Jordan
01-26-2018, 08:48 AM
That's kind of the opposite of what I was suggesting. What I was suggesting was changing the settings in old LW to match how 2018 is rendering it *, to achieve a more comparable render operation. And also as I noted, that wouldn't take anything away from the fact that 2018 is a great workflow improvement in that regard, even if the render time differences become smaller.

(Edit: * in the ~1 minute rendering)

That is exactly what I'm working on here this morning. I thought that would be a very interesting test and as close as an apples to apples comparison as we might get.

jakuzaa
01-26-2018, 09:39 AM
My test ! Core I7 6950 X - 20 Threads - 40 mn with LW 2018 - 1920 X 1080
HDRI - No GI
Happy but working with Octane in the studio and 40 mn in this format it's long............. Max 10 mn with Octane

139759

Max 10 min with Octane on what hardware exactly?

Asticles
01-26-2018, 09:41 AM
Wow, great scene! I suppose a couple of Titans and half of the Olympus gods... :P

THIBAULT
01-26-2018, 09:43 AM
max 10 min with octane on what hardware exactly?

6 titan x / 12 go

Nicolas Jordan
01-26-2018, 09:45 AM
Max 10 min with Octane on what hardware exactly?

Yes with Octane that is always the big question.

- - - Updated - - -


My test ! Core I7 6950 X - 20 Threads - 40 mn with LW 2018 - 1920 X 1080
HDRI - No GI
Happy but working with Octane in the studio and 40 mn in this format it's long............. Max 10 mn with Octane

139759

Very nice!

nagra
01-26-2018, 10:21 AM
My test ! Core I7 6950 X - 20 Threads - 40 mn with LW 2018 - 1920 X 1080
HDRI - No GI
Happy but working with Octane in the studio and 40 mn in this format it's long............. Max 10 mn with Octane

139759

My first posting hereÖ

This render is a good example for something i came along with while making some test renders.

Nobody else noting that renders with environment light and GI off result in renders with no light from down to top at all? Look e.g. at the ceiling or the down facing borders they are completely black at daytime, not really realistic.
In my own test renders I noticed that everything that faces exact downwards is rendered black with env light and GI off and only way around it is to set GI on. Am i the only one?

RebelHill
01-26-2018, 10:47 AM
It depends on the occlusion in the scene. If you have a ground plane which blocks light coming from an enviro's "southern hemisphere", then it cant light overhangs from below... but neither could BG GI (with the 1 bounce). In such a case you have a couple of options... If using BG GI, youd have to have 2 bounces, but with enviro light the first "bounce" is the light itself, so you now only need to add 1 GI bounce... You can add GI to your enviro light setup and its STILL faster to render than a "all GI" version would be. Option2, you could add 2 enviro lights... one lights the building but ignores the groundplane, the second lights the ground plane but ignores the building. This would be a less accurate solution, but depending on the situation, it may be passable enough, and would be faster still.

nagra
01-26-2018, 11:59 AM
... You can add GI to your enviro light setup and its STILL faster to render than a "all GI" version would be. ...

Not sure i got you right. I assume with bounces you refer to the # of diffuse bounces in the render properties.
In my test scene (a bridge on an asphalt-road like ground plane, which i lost btw because i forgot to save it) i increased the # of diffuse bounces to 3 without getting the result i was looking for. Seems the dark grey ground in that scene did not bounce back enough light. Anyhow with global GI on it worked as expected. Maybe i messed something up with the different settings but canít check as i donít have the scene anymore. I recall the scene had an env light and a distance light.

madno
01-26-2018, 12:15 PM
Also tried the scene. Did not care about hitting color etc. (also ignored that the house does not have a back wall).

Rendered on dual Xeon (32 threads (maybe six years old)):

GI-OFF_3m46sec.jpg
139771

GI-OFF_5m36sec.jpg
139772

GI-ON_6m20sec.jpg
139773

Put sky in textured env. and used env. light and a distant as sun
Materials are mostly BSDF with reflection.

Anyway, was just a test.

RebelHill
01-26-2018, 01:27 PM
Seems the dark grey ground in that scene did not bounce back enough light.

Of course it doesnt... its dark grey, so it can only bounce back "dark grey light".

samurai_x
01-26-2018, 05:26 PM
Wow, great scene! I suppose a couple of Titans and half of the Olympus gods... :P

With a fast renderer it won't take a couple of titans to do that kind of render. its a very typical quality at Evermotion using vray. Literally thousands of vray users can do that kind of render for the past 10 years. Almost doesn't get attention anymore because its so common for everybody.
https://evermotion.org/vbulletin/showthread.php?123828-Contrate-Exterior

gar26lw
01-26-2018, 07:20 PM
With a fast renderer it won't take a couple of titans to do that kind of render. its a very typical quality at Evermotion using vray. Literally thousands of vray users can do that kind of render for the past 10 years. Almost doesn't get attention anymore because its so common for everybody.
https://evermotion.org/vbulletin/showthread.php?123828-Contrate-Exterior

that’s the kind of test scene i’d like to see. a real comparison against vray.

gar26lw
01-27-2018, 02:54 AM
139788

perhaps a test with the latest modo beta. they have nightly builds.

http://forums.newtek.com/attachment.php?attachmentid=139788&d=1517046918

Asticles
01-27-2018, 05:41 AM
Also tried the scene. Did not care about hitting color etc. (also ignored that the house does not have a back wall).

Rendered on dual Xeon (32 threads (maybe six years old)):

GI-OFF_3m46sec.jpg
139771

GI-OFF_5m36sec.jpg
139772

GI-ON_6m20sec.jpg
139773

Put sky in textured env. and used env. light and a distant as sun
Materials are mostly BSDF with reflection.

Anyway, was just a test.

Much better the third.

nagra
01-27-2018, 03:17 PM
Of course it doesnt... its dark grey, so it can only bounce back "dark grey light".

Yes mostly besides the fresnel effect. But the color of the ground was not the reason.

I made more tests with the GI test scene provided by Exception. The distance light was replaced by a environment light.
Result: With GI off the engine does not calculate bounces no matter which value in the prefs is set for diffuse bounces.

First two render with 100% white walls (specular 0%), there is no difference between 1 and 3 bounces.

139802

Now with walls set to 100% specular and 100% roughness. Again no difference between 1 and 3 bounces.

139803

In short with GI off the engine works like a ray-tracer no light bounces and thats why its becoming fast.

Last but not least two render with GI on that clearly show the difference between 1 and 3 bounces. Now the engine works obviously as a path-tracer.

139804

Sorry for the low render quality but my other machine is busy.

RebelHill
01-27-2018, 03:26 PM
The distance light was replaced by a environment light.
Result: With GI off the engine does not calculate bounces no matter which value in the prefs is set for diffuse bounces.

Obviously not... because GI is needed for bounced light. However GI is also used to cast light from a backdrop/environment... which, when so done, counts as the first bounce. Use BG GI, 1 bounce, you get the same thing as enviro light. The difference is the enviro light is faster, and thus allows for fewer bounces.

nagra
01-27-2018, 03:49 PM
No doubt the environment light saves you 1 bounce and is therefore best choice for out-door render. For in-door portals or panel lights may give better results.

samurai_x
01-27-2018, 05:20 PM
thatís the kind of test scene iíd like to see. a real comparison against vray.

Nothing beats vray on cpu. The only competition was corona and that's been bought. :D

gar26lw
01-27-2018, 07:28 PM
Nothing beats vray on cpu. The only competition was corona and that's been bought. :D

this is a legitimate question..
so what’s the point in a new renderer? why not just lobby for vray in lw and contribute resources.

samurai_x
01-27-2018, 07:30 PM
People have asked for it for more than 10 years. CG isnt interested. It has to be a third party.

gar26lw
01-27-2018, 07:35 PM
People have asked for it for more than 10 years. CG isnt interested. It has to be a third party.

a third party that implements it or makes another renderer like it?

they made it for modo. too small a market?

Wickedpup
01-28-2018, 02:10 AM
One big difference is that within the Modo community Vray has been welcomed with open arms, while here it has been just as much "Vray? We don't need it" as it has been "Vray?Yes, please". But when LWG develops a new renderer everything is hunky-dory. Go figure.

samurai_x
02-05-2018, 11:57 PM
I've seen this render from different renderers. Probably originally from vray.
People are sharing it on fb.

LW 2015.3 + OCTANE
3 MIN / 880M / NO POST
https://scontent.fceb2-2.fna.fbcdn.net/v/t1.0-9/27459615_10155748472531311_2889388221243034498_n.j pg?oh=806d1ee19a2c042afcbcdc31fb09b8b3&oe=5B21D83B
lw octane scene.
https://drive.google.com/file/d/1a7ncVZpH0z3ns1EGheXbOrd77TQiEf4_/view


LightWave 2018 renderer
26 min, i7 950, 8 threads, noise filter, standard shaders, GI around 2min 20sec, the rest is antialiasing, 750x1000px, no PS
https://scontent.fceb2-2.fna.fbcdn.net/v/t1.0-9/27540472_1577483305672655_6178284457570668484_n.jp g?oh=17666ae6d97dbd71dc0037e3f6af8e4c&oe=5B1BC01B
lw 2018 scene
https://drive.google.com/file/d/1bOlxzbuKsY3ai1QN9Qg5Ep2t1r8Cwj7o/view



LightWave kray
5 min 25 sec
https://scontent.fceb2-2.fna.fbcdn.net/v/t1.0-9/27336877_1615491271850069_2868936519924174513_n.jp g?oh=7293cf8d33dab8b96e623fd22faf9c7f&oe=5B11A5A8

MarcusM
02-06-2018, 12:49 AM
In LightWave example SCENE file is for image below so you can replace:
- interploated, 53 min on i5 (4 threads), standard materials, no noise filter, no post

139957

jwiede
02-06-2018, 02:39 AM
LightWave kray
5 min 25 sec
https://scontent.fceb2-2.fna.fbcdn.net/v/t1.0-9/27336877_1615491271850069_2868936519924174513_n.jp g?oh=7293cf8d33dab8b96e623fd22faf9c7f&oe=5B11A5A8

For 5m render time, that's a really nice image from Kray (v2, no less).

It'll be interesting to see if someone can get LW2018 results as nice in an approx. similar timeframe (with blurry reflections, etc.).

jwiede
02-06-2018, 12:50 PM
Do you have the Kray2 version of the scene available for download? I'd like to see what MODO Vray can do with it, but MODO can't (yet) read LW2018 LWO/LWS. Converting from the Kray2 version should also be a bit easier than converting from the LW/Octane version -- if not, no worries, just figured I'd ask.

madno
02-06-2018, 11:30 PM
I've seen this render from different renderers. Probably originally from vray.
People are sharing it on fb.

LightWave 2018 renderer
26 min, i7 950, 8 threads, noise filter, standard shaders, GI around 2min 20sec, the rest is antialiasing, 750x1000px, no PS

https://drive.google.com/file/d/1bOlxzbuKsY3ai1QN9Qg5Ep2t1r8Cwj7o/view



Is there a reason why all surfaces are double sided or is it just the result of a default conversion to LW 2018?

samurai_x
02-07-2018, 08:38 AM
Do you have the Kray2 version of the scene available for download? I'd like to see what MODO Vray can do with it, but MODO can't (yet) read LW2018 LWO/LWS. Converting from the Kray2 version should also be a bit easier than converting from the LW/Octane version -- if not, no worries, just figured I'd ask.

Pretty sure the scene is publicly available somwhere with a vray version.

EBD3D
02-07-2018, 11:16 AM
140004

Lightwave 2018.0.1
Rendu : 20mn en 750*1000

Rayek
02-07-2018, 12:28 PM
Many issues with that render. Splotches, light leaking, moire patterns, noise, panel details washed out,...

I wouldn't be able to show this to a client.

http://i66.tinypic.com/2j1prit.jpg

EBD3D
02-07-2018, 01:10 PM
I agree with you but why not have done the analysis of the image kray too, it has many defects: the ground is vague, on the walls there are losses of detail, no reflection of the image of the painting on the surface of the table, etc ...
Now the Octane version of 3mn, the rendering is in AO ...
And the image of 26mn made in lightwave 2018, what is it made to make believe that lightwave can not make a realistic image in 26 minutes.
I agree that the Kray image is the most presentable but not for a client, me for a client I do a normal picture not a speed contest or saber everything.
I am also an octane render user so it's not a criticism of other rendering engines just that they are not flawless either.
For me the lack of lightwave is the part GI with the impactes that we have difficulty starting even with high parameters, otherwise the engine goes in the right direction.

Rayek
02-07-2018, 02:37 PM
Agreed, the kray image washes out panel details as well. Octane's AO seems to resolve that - I found that the left door's panels are relatively difficult to get to look right. And yes, for a client we'd reserve much more time to render the final presentable image. The jpg compression doesn't help either.

A more realistic comparison would perhaps be a render time of ~30 minutes, and then compare the render quality of the various render engines. But this would be a fair comparison only when done on the same hardware and semi-identical settings.

And so much can be achieved in post. Difficult to compare directly when some render engines apply post-rendering processing while or directly after rendering.

Having said that, I see some issues in the LW 2018 renders that I am unsure belong. The moire pattern in the table is odd (possibly introduced by the noise removal?) and the light leaking is problematic as well. Not sure if it's due to the model, although I tested this scene in other render engines, and none displayed the same odd corner and splotches. The built-in denoiser seems to be at fault for some of the defects, so it might be a good idea to remove noise in another application. I think that is definitely the case when we examine the ceiling medallion: the detail is completely washed out in all renders.

After all the testing in the past three weeks experimenting with rendering in LW2018, I have to say I think they might be on the right track, if not the track that I wished for in the past decade. It is what it is. This latest release isn't nearly enough to pull me back in - too much needs to be addressed as of yet, but I will keep an eye out and see where LW will be at in a year or two.

I do wish, instead of re-inventing the wheel, Newtek would have focused on a tight integration with an existing GPU/CPU path tracer render engine (such as AMD Prorender). One of the benefits could have been that Lightwave doesn't choke up while rendering as it does now, and they could have implemented [fill in the blanks]. Well, I mentioned this before. Old hat and stuff, right?

Could've, Might've, .... Anyway.

Learning more about the render settings in LW 2018 made my Cycles and Prorender image render times go down significantly while maintaining quality since I have an increased understanding of path tracer render settings in general now. So, it's been quite useful :-)

My trial runs out in 7 days. I might do some more testing, and then it's "toodles, and goodbye" from me once more - until we meet again, old Lightwave.

jwiede
02-07-2018, 03:17 PM
Having said that, I see some issues in the LW 2018 renders that I am unsure belong. The moire pattern in the table is odd (possibly introduced by the noise removal?) and the light leaking is problematic as well. Not sure if it's due to the model, although I tested this scene in other render engines, and none displayed the same odd corner and splotches.

I've seen more than a few cases now from both LW2018.0 & LW2018.0.1 where "solid" walls' corners using "clean" geometry (non-overlapping, walls have thickness, etc.) appear with odd "light leaks" -- though it's fairly obvious from the geometry what's happening isn't technically a "light leak". If you recall, there was a similar issue in the upper left corner of the earlier image everyone was reviewing with the window, mirror & glass balls, and rods/shelves sticking out. In that case, the geometry was "clean" (inside and outside of corner were topologically separated by measurable thickness on order of inches, etc.) as well.

I'm wondering if these "light leak" effects are just a larger-scale version of the odd "bright pixel" artifacts occurring in various pBSDF-surfaced matte/rough surfaces seen of late. It's almost as if, instead of a "hot pixel", sometimes the engine generates a stray "hot 1-pixel-wide edge segment" instead. It definitely seems to occur most visibly on interior-type renders in the shadowed corners of walls containing light sources, though it could just be that it happens elsewhere as well but isn't anywhere near as visible in those cases.

I've now seen the "hot pixel" phenomena enough to rate frequency "approaching common" with matte-/rough-type surfaces. This "hot edge"/"light leak" phenomena seems similarly frequent with interior shadowed corner situations (as stated). Whatever is causing these issues (whether related or not), they appear frequently enough to represent a significant issue for the new render engine.

I really hope Lightwave support/development has put a high priority on fixing these issues. The artifacts they cause are too visible to ignore, the level of noise reduction needed to eliminate the "hot pixel" artifacts typically has undesirable side-effects on texture quality, and the "hot edge" artifact variant appears resistant to noise reduction-based removal in general.

My trial expired, so my interest/ability to do further testing is now quite limited. I'm not buying the upgrade just for further render engine testing, so am taking a "reconsider purchase later" position for the time being. I'll continue testing simple issues with demo as time permits, but more elaborate testing and debugging just aren't feasible without save.

EBD3D
02-07-2018, 03:37 PM
Thank you Rayek for the exchange Rayek, for the table I look and keep you informed ;)

samurai_x
02-07-2018, 10:30 PM
The scene is there. Give it a shot. Maybe blender cycles cpu and gpu mode.

madno
02-08-2018, 12:44 AM
First, thanks for the scene.
After a first look I found some places to optimize. I think the scene was initially not made for engine testing.

140010

Nevertheless, I tried with it.
What I noticed in a lot of render test in the forum is that often there is no reflection on the walls. To my feeling renders look much better with it active.
The other one is that high interpolated GI smoothness washes out everything. Noise reduction does blur everything. Noise reduction seems to be useful only if the render is already very clean. But I also noticed that it takes time to render with reflection. Lot of samples needed to get an at least somehow nice output. Refraction (this other scene with glass ball in a room) and caustics are a total challenge (samples, samples and even more samples).
It seems, getting a perfect clean render with all the fancy stuff does not happen in a few minutes with LW (and maybe other non GPU renderers as well).

But we have the pass preview now to see where the problem is. I like that.

Examples still noisy (and with the geometry as it is).

14m39s_BF
140011

40m22s_BF
140012

THIBAULT
02-08-2018, 01:29 AM
Lw2015.3 + octane 3.07 / 6 x titan X
Pathtracing 12000 Samples
15 mn in 1500 X 2000 ! Octane is the LW 2018 future !

140013

madno
02-08-2018, 01:36 AM
Nice.

Would be interesting to know how long it would take with only two Titans. Do they scale linearly (6 times the cards = 6 times the speed)?
(geometry issues same like with LW native)

Edit:
Was there a version posted with the textures?

THIBAULT
02-08-2018, 01:41 AM
Yes, same version but i've changed render setting !
Yes, they scale linearly ! I'm going to test if you want !

Env. 00H45 with 2 TITAN X !

madno
02-08-2018, 02:01 AM
Ah, the Octane version came with textures.

Don't want to waste your time. If they are linearly 45m would be right. Maybe, if you are idle or bored, cross check it?

THIBAULT
02-08-2018, 02:21 AM
Yes, 45 mn !

madno
02-08-2018, 08:52 AM
Thanks.

lw_blender
02-08-2018, 10:03 AM
Hi guys, here is a quick blendercycles test i did to see how it could perform...

140019

13'30" on a single 1070GTX @ 750x1000

Testing now with LW 2015 and will post when it's finished.

samurai_x
02-08-2018, 10:43 AM
13 min on a single gpu with cycles? Thats pretty good.
Can you post the scene? Although the render looks a bit pale.

lw_blender
02-08-2018, 11:58 AM
13 min on a single gpu with cycles? Thats pretty good.
Can you post the scene? Although the render looks a bit pale.
Sure, here is the link to the file :
https://d.pr/f/dYv70Y (be careful, around 250MB!)

And yes, you're totally right about the overall look, I probably exaggerated a bit the white ambient look, nithing that couldn't be processed inside of blender though, given it has the compositing part integrated!

lw_blender
02-08-2018, 12:05 PM
And, as promised, here is a LW 2015 version of the scene, 31' on my bi-Xeon E5-2680v2 :

The raw render from LightWave

140023

And with a bit of tonemapping from the saved HDR

140024


Interpolated GI, DoF activated...

Lewis
02-08-2018, 12:09 PM
Lw2015.3 + octane 3.07 / 6 x titan X
Pathtracing 12000 Samples
15 mn in 1500 X 2000 ! Octane is the LW 2018 future !

140013

That looks great but it's pricey machine (esp today when GPUs has hard and expansive to buy due mining trend) for most users (especially LWavers). Do you have Octane scene so i can test, i think it might even ego lower to 10k samples and still be great quality. I'm looking forward to Octane 3.09 AI deoniser, that could solve all interior renders waiting and get it to very very fast times if denoiser proves to be usable as they showed in few tests.

Thanks

samurai_x
02-08-2018, 11:40 PM
Sure, here is the link to the file :
https://d.pr/f/dYv70Y (be careful, around 250MB!)

And yes, you're totally right about the overall look, I probably exaggerated a bit the white ambient look, nithing that couldn't be processed inside of blender though, given it has the compositing part integrated!

Thanks wil test.

Rayek
02-09-2018, 02:44 AM
Hi guys, here is a quick blendercycles test i did to see how it could perform...

140019

13'30" on a single 1070GTX @ 750x1000

Testing now with LW 2015 and will post when it's finished.

9 minutes on my 1080 & CPU (2.79 build with GPU&CPU simultaneous rendering).

A bit of tone mapping and NIK Sharpener Pro 3 - which does reveal the noise hidden in the image, and weak areas (table reflection).

Still, for 9 minutes not a terrible result, I'd say. At least the detail in the ceiling medallion is maintained.

*edit* on second glance & thought, I think it's too noisy. And detail is lost in the panels left.


http://i63.tinypic.com/27yy90m.jpg

lw_blender
02-14-2018, 02:54 AM
9 minutes on my 1080 & CPU (2.79 build with GPU&CPU simultaneous rendering).

A bit of tone mapping and NIK Sharpener Pro 3 - which does reveal the noise hidden in the image, and weak areas (table reflection).

Still, for 9 minutes not a terrible result, I'd say. At least the detail in the ceiling medallion is maintained.

*edit* on second glance & thought, I think it's too noisy. And detail is lost in the panels left.


http://i63.tinypic.com/27yy90m.jpg
Yes, the left panel, is, for an unknown reason (i didn't take the time to really see what was happening...), completely black in the first render i did.
I even added a small emission shader to see what is supposed to be the photo in the frame but it doesn't do it really...

As for the noise, did you use the integrated denoiser in blender, which is really impressive in term of ease of use and efficiency, at least compared to what LW 2018 seems to offer?

I have done a 3000x4000 test over the week-end on LW 2015 of this scene, which took 57h43' !!
But I used full Monte Carlo (interpolated) GI, with the directional rays enabled (doesn't make a difference in this scene, I think...), and let it cook while I was away.
I'll post the result later.

Have yet to see any impressive archviz render done with LW 2018 that couldn't be done in LW 2015, even in less time, don't know if someone did grasp the good settings and will show anything interesting?

gar26lw
02-14-2018, 05:56 AM
Have yet to see any impressive archviz render done with LW 2018 that couldn't be done in LW 2015, even in less time, don't know if someone did grasp the good settings and will show anything interesting?

ironically, i’ve been seeing a lot of renders in everything else.

lw_blender
02-16-2018, 03:32 AM
Here is the 3000x4000 render done in LW 2015, with all GI options turned on (Interpolated, still...), not sure it makes any difference really, except it took "quite a long time", 57h43' !!

As for now I see no reason for archviz people to upgrade to 2018 as the render engine doens't seem to make a big difference in term of speed an I feel the materials are now much more difficult to get an "artistic" look compared to 2015, while real physical engines, like Indigo or Thea Render are much more advanced in the lighting department or global illumination generally speaking...

Raw render form LW 2015, NO post process at all...

140153

UnCommonGrafx
02-16-2018, 04:50 AM
There is an hour and 31 minute render of this being shown on FB. https://www.facebook.com/photo.php?fbid=695844030806070&set=gm.1334028573369805&type=3&theater&ifg=1

There is a youtube posting of his process, as well.
Robert

tyrot
02-16-2018, 06:13 AM
57 hours ! this is insane ...OCTANE is all the way. I do not want to discourage anyone but .. i will write one more time.. JUST BUY a second hand 780 GTX and an OCTANE license (especially after denoiser sky is limit) Please do NOT HURT your business.

lw_blender
02-16-2018, 07:11 AM
57 hours ! this is insane ...OCTANE is all the way. I do not want to discourage anyone but .. i will write one more time.. JUST BUY a second hand 780 GTX and an OCTANE license (especially after denoiser sky is limit) Please do NOT HURT your business.
I know it's overkill, especially since adding "directional rays" (aka "caustics" in 2018) probably adds a lot of render time while having NO visible effect on the scene...

Anyway I wasn't at home last weekend so I thought I would let it cook as long as necessary... ;)
And don't forget it's a 3000x4000 render, I doubt Octane (which I own by the way but don't really have a use for it...) would "eat" that in 30 minutes either... :D
But you're welcome to prove me wrong, of course.

And as for faster render engines I've already shown a (very quick) blender Cycles render which took a little more than 10 minutes on a "poor", single GTX 1070 @ 750x1000 so you can also go to blender, which is free, and render a lot faster than in LightWave.


@Robert:
Thanks for the link (I'm not on FaceBook, never will be...), will have a look at how he used LW 2018 to see if there is any reason to upgrade, but I really doubt it by the look of it, seeing what is lost in upgrading to LW 2018 (IFW2 shaders, DP plug-ins, etc, etc...)

Rayek
02-16-2018, 12:15 PM
Here is the 3000x4000 render done in LW 2015, with all GI options turned on (Interpolated, still...), not sure it makes any difference really, except it took "quite a long time", 57h43' !!

As for now I see no reason for archviz people to upgrade to 2018 as the render engine doens't seem to make a big difference in term of speed an I feel the materials are now much more difficult to get an "artistic" look compared to 2015, while real physical engines, like Indigo or Thea Render are much more advanced in the lighting department or global illumination generally speaking...

Raw render form LW 2015, NO post process at all...

140153

I am a bit worried to see a 57 hour path tracer render result which STILL has visible noise, but, way more problematic, has obvious splotches in the shadows and ceilings. The panel lines are affected by this as well.

How can this be? Splotches in a 57 hour render? Did you turn on noise reduction? At first glance I thought that some kind of marble texture is applied to the walls and ceiling...

Is this due to a wrong setting?

Ztreem
02-16-2018, 12:57 PM
I am a bit worried to see a 57 hour path tracer render result which STILL has visible noise, but, way more problematic, has obvious splotches in the shadows and ceilings. The panel lines are affected by this as well.

How can this be? Splotches in a 57 hour render? Did you turn on noise reduction? At first glance I thought that some kind of marble texture is applied to the walls and ceiling...

Is this due to a wrong setting?

Its a lw2015 render with interpolated GI.

Rayek
02-16-2018, 01:18 PM
Duh! Should pay more attention.

pmwhite
02-16-2018, 02:33 PM
No I didn't use package scene because it usually crashes for me so I just did it manually. Package scene does work properly from time to time for me but I have found that it's not very stable.

Hello Strider, Thanks for the scene.
Did you report the Package crash issue, to Tech Support.
Cheers.

lw_blender
02-16-2018, 02:43 PM
Its a lw2015 render with interpolated GI.
Yep, that's it, LW 2015 render with Interpolated GI, brute force rendering, especially with "directional rays" (aka "caustics" in 2018) activated would have been even longer ! ;)

That said the 57h43' is for a 3000x4000 render AND would take at least 3 times less to render if I had optimized the settings to look as good but with render times in mind. :)
As I said I wasn't at home last weekend so I just decided to let it run and see the result.
But I must admit I thought it would be better for a 57h render... Blender would run Cycles (pun intended...) around it for that kind of scene.

I might try Kray 3 RC5 as it's just released to see how it performs.

jwiede
02-16-2018, 03:33 PM
I might try Kray 3 RC5 as it's just released to see how it performs.

There are Kray 2 numbers a bit further up in the thread, and they're pretty impressive. I agree, it'd be interesting to see how Kray 3RC5 compares.

rustythe1
02-17-2018, 09:08 AM
Well i decided to get in on the action, i approached the scene completely differently (so i will guess someone will pick holes for realism or things that they might not see physically correct! but all the original renders are far from physically correct and more from an artistic view, the main problem is that there is only one window and an open door in this room so the area close to the camera should be really dark and would be if you took a photo with no flash)
anyway, on my 5 yr old i7,
first render at the original scene size, 85 seconds!
140161
2x size around 5.5 mins
140162
extra large around 11 mins
140163

140164

i know some people have pointed out previously that the dark shadows at the end of the room are light leaks as they have what looks like light stripes going down the wall, its not the case, the shadows are from the curtains and light is coming from behind these as they are mounted away from the wall, so from what i can see the other renderers seem to be missing this, i.e. the light from within the room (of which there should technically be none) is overpowering the light coming from the window.

THIBAULT
02-17-2018, 10:56 AM
And your settings please ! Kray ?

Asticles
02-17-2018, 11:07 AM
This is only direct lighting, with AO, isn't it?

lw_blender
02-17-2018, 11:28 AM
Well i decided to get in on the action, i approached the scene completely differently (so i will guess someone will pick holes for realism or things that they might not see physically correct! but all the original renders are far from physically correct and more from an artistic view, the main problem is that there is only one window and an open door in this room so the area close to the camera should be really dark and would be if you took a photo with no flash)
anyway, on my 5 yr old i7,
first render at the original scene size, 85 seconds!
140161
2x size around 5.5 mins
140162
extra large around 11 mins
140163

140164

i know some people have pointed out previously that the dark shadows at the end of the room are light leaks as they have what looks like light stripes going down the wall, its not the case, the shadows are from the curtains and light is coming from behind these as they are mounted away from the wall, so from what i can see the other renderers seem to be missing this, i.e. the light from within the room (of which there should technically be none) is overpowering the light coming from the window.
Cool renders!
...Apart from the fact that this scene isn't, imho, here to show artistic skills or points of vue on how it should look like (for this, kudos to the creator of this scene!), but more on how the render engines can behave with this scene and their internal GI capabilities.
Otherwise, we can also bake the GI on the objects and run a 60fps Unreal simulation, now THAT will be fast! ;)

Anyway, I'm not the creator of the scene, nor am I on this forum since a long time so i'll let everyone see what there is to conclude regarding those renders. :D

madno
02-17-2018, 12:24 PM
Well i decided to get in on the action, i approached the scene completely differently (so i will guess someone will pick holes for realism or things that they might not see physically correct! but all the original renders are far from physically correct and more from an artistic view, the main problem is that there is only one window and an open door in this room so the area close to the camera should be really dark and would be if you took a photo with no flash)
anyway, on my 5 yr old i7,

...

i know some people have pointed out previously that the dark shadows at the end of the room are light leaks as they have what looks like light stripes going down the wall, its not the case, the shadows are from the curtains and light is coming from behind these as they are mounted away from the wall, so from what i can see the other renderers seem to be missing this, i.e. the light from within the room (of which there should technically be none) is overpowering the light coming from the window.

Do you mind sharing more info about your approach?

By the way: I just found your downloadable models - COOL (loved the original Starship Troopers).

rustythe1
02-17-2018, 02:27 PM
No AO, its just lights, i set the scene up as if i was setting up a photo, it has a lot of tweakability and to get cleaner renders you can just up the light samples as the only place you really get any noise is in the rough surfaces like wood, there are 4 area lights and spherical light that kind of acts like AO, only the lights at real light sources affect spec, the other lights are acting like diffused camera flash units, so large with no spec effect,
if you put them side by side with Kray etc, i much prefer the 85 seconds render over the 5 mins in kray, the kray and octane ones are far more washed out and show a lot less detail and solidarity, in some places you completely lose the wall mouldings, don't know if that's user error or artistic but i see a lot of washing out and not much dark separation, which also some of my customers used to complain about if i tried to go in that artistic direction as it often shows in Vray renders, oddly some people find it adds more realism, but most who i work for want to see all the detail and i think the LW engine has a very good balance on realism and still being able to show a very dynamic range of detail, here are 3 side by side, left, lw how i have it now at 85 seconds (still playing around with lights so it looks a little different now) samurais 5 min render in kray and then mine again with the lights upped in case you want that washed out look, yet you can still see all the detail in the mouldings, and still see the image maps in the two paintings, lights and curtains but have almost vanished to black in kray,
140170
so i think from here there is no argument for speed for LW2018, its all down to personal taste, and i can tell you if i showed these 3 to one of my customers they would not pick the one in the middle or the 3 min octane one over the two LW ones (although they would also moan at me that the third was too washed out any way)

rustythe1
02-17-2018, 03:19 PM
so it just looks like this, none of the area lights at portal, just set to normalize, the two at the rear and the spherical are about 1/3 the value of the front ones, and only set to affect diffuse, i think i had to set the light multiplier to 10 in render properties but i may have my colour space set wrong as i didn't check it as its the imported octane scene and i don't have octane in 2018
140171

lw_blender
02-17-2018, 05:18 PM
@Rustythe1 :
Again I think you completely miss the point of this thread, which isn't how to optimize a scene with fake lights...
... As you cannot always fake everything, or not to the point a good render engine would take you with its better algorithms.... Otherwise we wouldn't have seen such dramatic progress in 3D renders over the past 20 years!

This scene is really light and easy, so easy to fake with well placed lights, and it's a great thing to be able to render fast that kind of scene really fast with "GI lights", I've had my share of scene optimization over the last 20 years myself and I surely appreciate that kind of effort.
Anyway that's not what help a render engine to progress in ease of use and speed with modern techniques, which is, I think, the whole point of this thread.

We're not talking artistic here, the gallery is here for that, with (preferably?) original creations, we're talking about settings to let the new GI engine of LW 2018 show its full potential, and how it compares to other render engines, like Kray or Octane or even LW 2015.
Would be great to see a LW 2018 render test of this scene (or any other, more complex preferably) which shows how cool the new GI engine can be.

Mind you, Gerard Strada showed a great deal of years ago (possibly more than 10 years now?) how you could fake GI with gradients ! It is/was a great technique as well, but it can't be used for every type of scene, sometimes you just have to let the render engine do its magic, that's also why render engines are constantly improving, let's hope LW 2018 is heading in the right direction as well. :)

rustythe1
02-17-2018, 05:46 PM
no, i didn't miss the point, the first post was to show if you can get comparable/similar or better results in lw2018 in the same or less time from lw2015, which i think i have shown, can you point out where in the image GI would make a substantial difference to the final image as compared to Kray and octane?
140172
sorry if text sounds out of tone, its a genuine statement as in is there any lack of realism in the left and right images compared to Kray and octane or even the 2015 on the previous page, 85 seconds compared to hours (up to 57 hrs i think from previous pages) answers part of the original question of this thread, can it be done quicker?

gar26lw
02-17-2018, 06:26 PM
it’s very nice rustythe1 but most want to press the button marked “make cool render” then go have some lunch. ;)

i know id rather use a render engine that did that over one where i have to change every setting and setup lots of lights to match the “make cool render” button. time is money.

samurai_x
02-17-2018, 06:46 PM
W
first render at the original scene size, 85 seconds!
140161



Well it does look like a render that is 85 secs long.
It looks very much like a scanline render from 3dmax in the 90's which renders fast back then, too.


Fwiw, these render you put side by side are horrible in todays standards.
140170

lw_blender
02-18-2018, 02:46 AM
no, i didn't miss the point, the first post was to show if you can get comparable/similar or better results in lw2018 in the same or less time from lw2015, which i think i have shown, can you point out where in the image GI would make a substantial difference to the final image as compared to Kray and octane?
140172
sorry if text sounds out of tone, its a genuine statement as in is there any lack of realism in the left and right images compared to Kray and octane or even the 2015 on the previous page, 85 seconds compared to hours (up to 57 hrs i think from previous pages) answers part of the original question of this thread, can it be done quicker?
Well, if you don't see the "subtle" (?) differences between the renders on a scene that is so simple then I don't think I can do anything for you, I guess...

Again, like I said, this scene is still "easy" to fake, not all archviz renders can benefit as much as this scene from faking GI with other techniques, it has to have a great (and fast, that's the whole point of this thread...) "GI engine", that is reliable in every circumstances.
... And there are loads of them already, I wish LW 2018 is/will be part of those render engines able to sort out complex GI scenes reliably and in a reasonable amount of time, that's what we are looking for. :)

rustythe1
02-18-2018, 11:13 AM
ok, we can do it your way, here are 6 for comparison,
140178
it shows me kray loses a lot of detail in the mouldings, and also the blacks, see the bin and curtains, the reflection in the table is all wrong as its not reflecting the reflection on the painting, harsh floor textures and not reflecting the window frame,
the octane one suffers similar black problems, and also similar reflection problems, the painting on the table is only reflecting the image bellow the glass and not the reflection on the glass, curtains are showing a bit more but still a loss of detail, and again harsh floor textures and not reflecting the window frame, mouldings are a lot better so a lot more white differential,
cycles, shows the blacks a lot better especially in the bin, although table reflection still off in this one, floor textures are nice but maybe not shiny enough compared to everybody else's (obviously just a surface setting though) glass in the vase is not working
LW no gi, obviously fast and clean so good when your in a hurry or need to show very fine details like the mouldings, to me it looks good even if anti lightwave say otherwise,
LW Gi only, seems to be a transparency bug as the vase and two pictures are not rendering correctly (although they do in interpolated) still a bit of noise after 11 mins but not as much as in the Kray wood and far left door,
LW Interpolated, most of the noise is gone, but the splotches on the left are a difficult one, but then i only let it go for 4 mins, so you could triple the settings and run as long as the blender one,
overall i think the floor, table curtains and paintings are much better in all 3 of the LW2018 renders but the octane one shows better whites (although all the shadows are missing from the curtains on the walls)

lw_blender
02-18-2018, 12:15 PM
ok, we can do it your way, here are 6 for comparison,
140178
it shows me kray loses a lot of detail in the mouldings, and also the blacks, see the bin and curtains, the reflection in the table is all wrong as its not reflecting the reflection on the painting, harsh floor textures and not reflecting the window frame,
the octane one suffers similar black problems, and also similar reflection problems, the painting on the table is only reflecting the image bellow the glass and not the reflection on the glass, curtains are showing a bit more but still a loss of detail, and again harsh floor textures and not reflecting the window frame, mouldings are a lot better so a lot more white differential,
cycles, shows the blacks a lot better especially in the bin, although table reflection still off in this one, floor textures are nice but maybe not shiny enough compared to everybody else's (obviously just a surface setting though) glass in the vase is not working
LW no gi, obviously fast and clean so good when your in a hurry or need to show very fine details like the mouldings, to me it looks good even if anti lightwave say otherwise,
LW Gi only, seems to be a transparency bug as the vase and two pictures are not rendering correctly (although they do in interpolated) still a bit of noise after 11 mins but not as much as in the Kray wood and far left door,
LW Interpolated, most of the noise is gone, but the splotches on the left are a difficult one, but then i only let it go for 4 mins, so you could triple the settings and run as long as the blender one,
overall i think the floor, table curtains and paintings are much better in all 3 of the LW2018 renders but the octane one shows better whites (although all the shadows are missing from the curtains on the walls)
Thanks for your testing, much appreciated! :thumbsup:

Now THAT is some information! :rock:

As for your "non GI" render, it's not about being "anti-LW" (at least not for me, I'm using it since 23 years now...) as much as it about comparing engine with (almost) equivalent settings and techniques engaged, that is what is interesting here. I guess other software have their shortcuts as well. :)
Anyway, that's a good base for comparison, so thanks again for contributing!

rustythe1
02-18-2018, 12:32 PM
Anti lightwave wasn't aimed at you, there is just a group of people that grace the forums but appear to only be here to put lightwave down and promote all other software as all other software seems to be faster and better than lightwave, and I'm not pointing fingers but it seems we have a lot of render experts (not even pointing directly at this thread) but they seem to not be able to work quick render times (although it seems they don't even use lightwave, have no interest in it, or just here to point out the failings of it) anyway, that's my vent for the evening over (hopefully) but you can expect the barrage to follow when the time differences catch up! :)

OjN
02-18-2018, 04:36 PM
Thanks rustythe1! Nice test.
This my weekend test scene, Krita post, with GI and LW18=10(bakeR)+10min(render) in and old slow +5years AMD...

140182
I use 2 sets of lights, one for bake the radiosity solution (try to speed up it) and other for final render (with only the necessary lights). I tried to fix it in post the low quality gi...a little whashed:) Fireflies and reflections need some more samples, material work and tweaks...

jwiede
02-18-2018, 10:35 PM
ok, we can do it your way, here are 6 for comparison,
140178
...
overall i think the floor, table curtains and paintings are much better in all 3 of the LW2018 renders but the octane one shows better whites (although all the shadows are missing from the curtains on the walls)

The LW2018 version of the floor seems too "mirror-like". If you look at the FStorm-targeted original (see here (http://zchen.ca/interior-fstorm/)), the original intent seems to fall somewhere between Kray and Blender. In the FStorm version, the wood grain depth, edging detail, and texturing are all quite visible, while the reflections from the window are nowhere near as clear or "prominent", and _much_ blurrier than they are in any of your LW2018 renders.

rustythe1
02-19-2018, 03:31 AM
That's fair enough, i had not seen the original source as it was the edited octane one from FB and most the renders i have seen show it more glossy, i would assume there was supposed to be some sort of normal map or normal from image that was creating displacement in the original, i only presented the texture images as is, so the point i was making is if the octane and kray renders here have done the same they are showing a much harsher image than that of the original texture images (this could be intentional or it could be the way the renderer works), the same reason my table is white, the image provided for that is a white marble, the curtains are grey, but he must just be using these as multiply rather than direct maps.

Asticles
02-19-2018, 11:46 AM
My first attempt. Noisy, but it is a matter of time.

28m brute force on i7 laptop.

Regards

140194

lw_blender
02-19-2018, 06:17 PM
Ok guys, last test for me, this time with Kray 3 RC5, the latest, which I don't really know well for now, so there is probably room for (big) optimization...

Anyway, i have done 2 renders, one in 750x1000 and the second one in 1500x2000 to better show the overall quality of the render engine, with more pixels.
And obviously Kray 3, when it will be finished (hopefully sooner rather than later, given it's been for so long time in development...), will deliver, it's suite impressive to see it crunch that type of scene as if it was nothing!

Those renders are without any irradiance cache, and they show a little tiny bit of noise in the GI solution, which is very acceptable given there is NO noise reduction in action!
provided as is, without any post-production done.

Render times :
@750x1000 : 9'21" on my i7/6700K hackintosh / 3'34" on my Windows 10 Bi-Xeon E5-2680v2
@1500x2000 : 34'58" on my i7/6700K hackintosh / 13'18" on my Windows 10 Bi-Xeon E5-2680v2

140198

140199

jwiede
02-19-2018, 06:26 PM
Those renders are without any irradiance cache, and they show a little tiny bit of noise in the GI solution, which is very acceptable given there is NO noise reduction in action!
provided as is, without any post-production done.

Render times :
@750x1000 : 9'21" on my i7/6700K hackintosh / 3'34" on my Windows 10 Bi-Xeon E5-2680v2
@1500x2000 : 34'58" on my i7/6700K hackintosh / 13'18" on my Windows 10 Bi-Xeon E5-2680v2

140198

140199

Considering those are from a "pre-release" CPU-based render engine without noise reduction, they're quite good quality, and the times are really good. The noise is really minimal, and the floor came out excellent -- it looks quite close to the FStorm version's floor in reflectivity and finish, and there's a really nice "richness" and depth to the color and texture. The panel and medallion details are also showing well, as are the curtain shadows (and the curtains' texturing also looks nice).

One of the best overall thus far, IMO. Now if they could just get Kray3 finished and released!

gar26lw
02-19-2018, 08:38 PM
so is Kray 3 the winner ?

samurai_x
02-19-2018, 09:55 PM
so is Kray 3 the winner ?

Its a good render with no noobish overbrights and good gradiation of light. Doesn't look faked, nothing too washed out, nothing too dark.

samurai_x
02-19-2018, 10:02 PM
The LW2018 version of the floor seems too "mirror-like". If you look at the FStorm-targeted original (see here (http://zchen.ca/interior-fstorm/)), the original intent seems to fall somewhere between Kray and Blender. In the FStorm version, the wood grain depth, edging detail, and texturing are all quite visible, while the reflections from the window are nowhere near as clear or "prominent", and _much_ blurrier than they are in any of your LW2018 renders.

Wow those renders are too bright. I guess he didn't go for beautiful renders. Just crunching the scene for benchmark.
The Redshift got the cleanest result with the least render time obviously. Vray had the better gi and contact shadows.
7min vs 45min with very similar looking images I would go for Redshift.

http://i1.wp.com/zchen.ca/wp-content/uploads/2016/12/1609_full_001.jpg?zoom=1.1041666269302368&w=1180

Rayek
02-20-2018, 02:03 AM
Considering those are from a "pre-release" CPU-based render engine without noise reduction, they're quite good quality, and the times are really good. The noise is really minimal, and the floor came out excellent -- it looks quite close to the FStorm version's floor in reflectivity and finish, and there's a really nice "richness" and depth to the color and texture. The panel and medallion details are also showing well, as are the curtain shadows (and the curtains' texturing also looks nice).

One of the best overall thus far, IMO. Now if they could just get Kray3 finished and released!

The reason for the nice looking panel detail: the light angle is very different from the previous renders. I tested this myself, and with that light angle washed out panels are avoided. So not quite comparable lighting conditions result in very different looking details.

Exception
02-20-2018, 02:23 AM
I'd like to try this scene... I got the 2018 version download from earlier, but it doesn't seem to include textures. Can anyone share a scene with everything in it for testing, or point to the right link I maybe missed? thanks!

MichaelT
02-20-2018, 02:48 AM
Why on earth are you using V-Ray RT as reference? It isn't production worthy, and not finished. It doesn't render scenes correctly all the time.

lw_blender
02-20-2018, 04:24 AM
The reason for the nice looking panel detail: the light angle is very different from the previous renders. I tested this myself, and with that light angle washed out panels are avoided. So not quite comparable lighting conditions result in very different looking details.
Yep, you're right, so I redid the "same" render without the direct sunlight, to be as near as possible to the original scene.

Here is the result, render time was, @ 1500x2000 (didn't bother to render @750x1000 this time...) : 33'21" on my i7/6700K hackintosh / 12'41" on my Windows 10 Bi-Xeon E5-2680v2
Don't forget that, as John said, it's still a beta of this render engine (hopefully, one day it will be "finished" and really released!), and I'm far from being a master using it.
I'm sure Janusz (THE Kray master!) would render the same image with even better quality in less time...

Anyway here is the result, again straight from the engine, NO prost-production at all :

140202

- - - Updated - - -


I'd like to try this scene... I got the 2018 version download from earlier, but it doesn't seem to include textures. Can anyone share a scene with everything in it for testing, or point to the right link I maybe missed? thanks!
Here is my take on LW 2018 :
https://d.pr/f/UCgXlA

Far from being optimized as I didn't really know what I was doing when trying to optimize the scene but it's a start if you want to play with it. :)

artzgo
02-20-2018, 05:35 AM
Kray 3 RC5

artzgo
02-20-2018, 06:34 AM
https://www.dropbox.com/s/1xitmai3hrg1a9m/K3-B.jpg?dl=0

https://www.dropbox.com/s/0y88hjbfrl0o1m5/K3-C.jpg?dl=0

MichaelT
02-20-2018, 07:01 AM
That is among the better renders I've seen so far :) Don't know anything about the Kray renderer though.

artzgo
02-20-2018, 07:02 AM
with 1 Area Light on the window
https://www.dropbox.com/s/9mlc1o71ecpn8ot/K3-with-Area-Light-1.jpg?dl=0

gar26lw
02-20-2018, 07:22 AM
very cool. how long?

artzgo
02-20-2018, 07:30 AM
night ver. only lumi light in the lamps
https://www.dropbox.com/s/gvvks0ovdb3jpdz/K3-night-ver.jpg?dl=0

- - - Updated - - -


very cool. how long?
you have a info bar on the bottom of the images

jwiede
02-20-2018, 07:55 AM
with 1 Area Light on the window
https://www.dropbox.com/s/9mlc1o71ecpn8ot/K3-with-Area-Light-1.jpg?dl=0

Yep, that's another rather nice one from Kray3rc5. Esp. given the time, lack of noise, and the richness in the floor (and lighting in general).

Kray3 looks like it'll be great for interiors, just as Kray2 was in its day.

rustythe1
02-20-2018, 07:56 AM
isn't the tread getting derailed again? each of these threads keeps turning into sales pitches for every other render engine (no offence to any of you showing them as they are of interest, but maybe not the subject of this thread) when the overall question of this thread was 2015 base LW against 2018 base LW, not how it goes against blender, octane, Vray, Kray, unreal, the guy just wants to know if he can render the same image in his new copy of 2018, which is why in the main i have been showing the less realistic ways as your trying to compare a more realistic lighting engine to less realistic one (note i said lighting not GI) as the old GI engine is quite comparable to the new light engine. and everyone seems intent on picking holes in the realism of purposefully less realistic renders against 2018 equivalent engines rather than comparing them only to 2015 version.

lw_blender
02-20-2018, 08:06 AM
isn't the tread getting derailed again? each of these threads keeps turning into sales pitches for every other render engine (no offence to any of you showing them as they are of interest, but maybe not the subject of this thread) when the overall question of this thread was 2015 base LW against 2018 base LW, not how it goes against blender, octane, Vray, Kray, unreal, the guy just wants to know if he can render the same image in his new copy of 2018, which is why in the main i have been showing the less realistic ways as your trying to compare a more realistic lighting engine to less realistic one (note i said lighting not GI) as the old GI engine is quite comparable to the new light engine. and everyone seems intent on picking holes in the realism of purposefully less realistic renders against 2018 equivalent engines rather than comparing them only to 2015 version.
Kray IS one of the few render engines available for LightWave (with Octane, that's the two of them remaining, as far as I know?) and is available for LW 2015 as well. In my case I'd rather keep my LW2015 copy and use Kray3 for archviz renders (which is the purpose of this thread) instead of getting a copy of LW2018 which doesn't show dramatic improvements in the GI calculation department.
... Which is fortunate I admit as I already own a copy of Kray3, at least of RC5 as the final release has yet to come up.

But I agree with you that all the tests done with other illumination conditions are not very on purpose. I did one myself, while exploring the possibilities of Kray3 and apologize for that as it didn't show what we are looking for here: Is LW2018 up to the task for GI archviz renderings? In term of realism or speed and ease of use?
My take on it would be that it could be in the future, but for now it isn't quite ready yet, unfortunately.

gar26lw
02-20-2018, 08:10 AM
richard yot on modo forum “I could of course create a more up to date tutorial for rendering interiors, it's just that I'm not sure how much value there would be there for people, when just hitting F8 gets you 99% of the way there. I've spent a long time researching the ins and outs of the Modo render engine, and got to know all the settings intimately, but I have to say I don't miss the days of tweaking sampling settings and doing test renders, Preview saves a lot of time compared to using F9”

press button, go for lunch ;). i hope we see a return of fprime style rendering in lw soon.

jwiede
02-20-2018, 08:12 AM
isn't the tread getting derailed again? each of these threads keeps turning into sales pitches for every other render engine (no offence to any of you showing them as they are of interest, but maybe not the subject of this thread) when the overall question of this thread was 2015 base LW against 2018 base LW, not how it goes against blender, octane, Vray, Kray, unreal, the guy just wants to know if he can render the same image in his new copy of 2018, which is why in the main i have been showing the less realistic ways as your trying to compare a more realistic lighting engine to less realistic one (note i said lighting not GI) as the old GI engine is quite comparable to the new light engine. and everyone seems intent on picking holes in the realism of purposefully less realistic renders against 2018 equivalent engines rather than comparing them only to 2015 version.

Well if you really want to go back to what the OP requested, his precise quote was: "Can this scene be rendered in 2018 at or very close to the same speed as in 2015 while still retaining the overall quality in areas such as AA, lighting, GI and reflection blurring?" Those renders you produced did not actually address the OP's request either, contrary to what you're suggesting.

OFF
02-20-2018, 09:38 PM
Nothing special just test with Kray3_RC5. 2xXeon's with 48HT.

140216

samurai_x
02-21-2018, 04:28 AM
People are waiting for a decent lw 2018 render of that interior. I might do a lw 2015 version.
Nobody's fault the decent ones came from another renderer.
The original scene from fstorm equally sucked. But I guess it was just benchmark stuff.

gar26lw
02-21-2018, 04:51 AM
People are waiting for a decent lw 2018 render of that interior. I might do a lw 2015 version.
Nobody's fault the decent ones came from another renderer.
The original scene from fstorm equally sucked. But I guess it was just benchmark stuff.

yeah do one man. id be keen to see it.

Chris S. (Fez)
02-21-2018, 05:15 AM
press button, go for lunch ;). i hope we see a return of fprime style rendering in lw soon.

Yup.

Exception
02-21-2018, 12:49 PM
Here is my take on LW 2018 :
https://d.pr/f/UCgXlA

Far from being optimized as I didn't really know what I was doing when trying to optimize the scene but it's a start if you want to play with it. :)

Thanks! I'll get on it.

madno
02-21-2018, 04:56 PM
LW 2018: 1h 21m

Unfortunately aliasing from harsh contrast here and there.

140229

jwiede
02-21-2018, 05:20 PM
LW 2018: 1h 21m

Unfortunately aliasing from harsh contrast here and there.

Yeah, LW2018 engine's AA seems prone to that (inadequate anti-aliasing of high-contrast gradients).

Anyone tried any really fine-detail line features, like black power lines at distance against light sky? Is it possible to apply enough AA to rid the aliasing without them turning into blurs (as seems to happen with high-contrast textures), without putting times through the roof?

lw_blender
02-22-2018, 02:26 AM
LW 2018: 1h 21m

Unfortunately aliasing from harsh contrast here and there.

140229
Thanks for your test!

Hard to see where GI is actually, as it seems like a non-GI render?
Too few bounces perhaps?

As an example my Kray3 renders have 40 bounces! (the advantage of photon/light mapping, setting a high number of bounces doesn't skyrocket the rendering times!)
I know the path tracers will always set that to a lot less bounces but here it seems like there is none or 1 at most...

Anyway, the blurry reflections are beautiful, without noise and that is a good thing for LW 2018. :thumbsup:
The walls are too reflective however, like they are made of plastic more than of plaster.

Again thanks for your test!
What kind of machine do you render on?

madno
02-22-2018, 05:09 AM
Note: All this information is from a hobby user standpoint. I don't know if my conclusion are correct. I admit, I am not able to get a good render in a few minutes time.


Thanks for your test!

Hard to see where GI is actually, as it seems like a non-GI render?
Too few bounces perhaps?

As an example my Kray3 renders have 40 bounces! (the advantage of photon/light mapping, setting a high number of bounces doesn't skyrocket the rendering times!)
I know the path tracers will always set that to a lot less bounces but here it seems like there is none or 1 at most...


Oh contraire - a hell number of GI interpolated samples in this one. E.g. you can still see blotchies in the left door tunnel.
I dare to say, it is a good sign if a spectator does not immediately notice that "THERE IS GI".



Anyway, the blurry reflections are beautiful, without noise and that is a good thing for LW 2018. :thumbsup:
The walls are too reflective however, like they are made of plastic more than of plaster.

Yes, agreed.

A little more roughness this time:

1h31m
140236


Again thanks for your test!
What kind of machine do you render on?

Dual Xeon E5-2687W @ 3.1 GHz (32 threads)
Ram 64 GB
HDD Samsung SSD 8XX something.
System is about six years old, SSDs came in later - waiting for Threadripper V2 in summer :)

NOTE to all testers:
If you have played with the scene at very low sample and AA settings and still got a slow VPR and Render, than this is the cause of it.

I noticed that objects might have individual GI settings (select object -> "p" for properties)

The ones I found in the scene left room for optimization.

E.g. the Room
140237
my settings

Primary / Secondary
Originally Primary Rays were 400, Secondary 200.
I played a lot and came to the conclusion that in this scene Secondary Rays increase render time consideribly, but do not really help. In the end without further testing I decided to use a lot of primary but only few secondary.

Missing Sample Rays:
Tried with high and low values but with no conclusion. So, I set it to 1.

Minimum:
Oh yeah, that's a nice one.
The lower the value the more detailed is the GI in fine detail areas (especially stucc).
But that costs. 1 is best, but rendertime goes up (not meassured the time but it felt like 50% to 100% for GI pass)

Maximum
No real conclusion. So, something in the middle.

Angular Tolerance
Another nice one.
The lower the value the better the details (contrast) in the stucc. Higher values = more washed out result.
15 looked nice to me, but of course - longer render times.
The effect is subtle. You might go higher without a "crying loud" visual effect.

The global GI looks like this:

140237

You'll notice that Sample Backdrop and ISBG is off.
i think, there is no relevant backdrop information (only white, no HDR etc.). So, no need to sample something that is not there. The light anyway comes from the two area lights in front of the windows (no Portals, because, like with GI, there is no relevant backdrop worth to sample. And because the lights are directly in front of the windows, the samples anyway go into the room. As i understand it, in this sceneario Portals just increase render time with no benefit).
ISBG: frankly speaking, I did not see a benefit of this either but it increased render time. So, off.

Interpolation Softness:
I felt that the default 50 is smoothing to much. Sure, it saves render time because one can lower the GI settings but I didn't like the blurry look.

Ah I forgot, all materials (except from the light blocker walls) have reflection.

And I loaded the room in modeler and hit "m" for merge points and "Unify Polys".

And the artifact in the middle of the light bulb is because of the geometry.

And so on, and so on.

There would me much more to do, to make this a good scene and good render.

One more - Blown Out Floor:
I was not able to lower the reflectance value in the floor planks BSDF material.
Even nodes did not help. It seems to me, once reflectance is on, it can't be reduced below a value only know by the newtek devs.
Maybe it is because of phyisical accuracy?
To avoid that nasty blown out effect seen in most of the renders, only very high roughness would help with BSDF. So I decided to fall back to simple Delta.

140238

jwiede
02-22-2018, 05:24 AM
Whatever you did, those are the best-looking versions of the floor yet from LW2018. Not quite as rich as lw-blender's Kray3rc5 one above (http://forums.newtek.com/showthread.php?155830-LW-2015-vs-LW-2018-Arch-Viz-render-testing&p=1537805&viewfull=1#post1537805), but definitely the best floor so far from LW2018!

I see what you mean with pBSDF reflectance, that feels like a bug (another pkg's "principled" material, based off same Disney source, doesn't exhibit the issue which suggests it isn't intrinsic to the shader). I'm guessing the devs just need to tweak min reflectance down for pBSDF, if I get a chance I'll try to file a bug later, but suggest you do so as well.

madno
02-22-2018, 02:14 PM
Thanks jwede:)

Here is an example after Affinity Photo.

Orig Render - Post

140260

madno
02-22-2018, 02:52 PM
And another one.

33m 29s
Brute Force / no reflection on room and doors

140261

jwiede
02-22-2018, 04:41 PM
And another one.

33m 29s
Brute Force / no reflection on room and doors

140261

There's something odd going on with the curtain shadows against the upper corners of the back (window) wall in that one. See the odd light vertical stripe present in both sides' curtain shadows?

madno
02-22-2018, 04:50 PM
Good eyes you have.

It's likely the light shining through the smal gap between the curtain and the window. Thought first it's like leak, but nope, don't think so.

Edit: and I moved the light more near to the window opening.

madno
02-22-2018, 05:16 PM
Another one (still same overall lighting - don't want to change in the middle of testing and understanding the renderer):

BF again with some adjustments. Reflectance is on again on the walls and doors, but less obvious this time (higher roughness value). I like the subtle depth it gives compared to the renders without reflection.

41m 42s
(by the way; if you compare the render times with the ones from lw_blender, keep in mind he is running 48 threads @ 3.5 GHz while I use 32 threads @ 3.1 GHz)
EDIT: Mistake on my side "his 33'21" render was done on a single i7/6700K" not on his dual Xeon EDIT END.

140263

Came out usable (except from the light bulb - (tried subsurface but did not set enough samples for that) and right painting (reduced refraction samples to much this time))
Noise Reduction gave it a final bit of further cleaning.

140262

jwiede
02-22-2018, 05:48 PM
Good eyes you have.

It's likely the light shining through the smal gap between the curtain and the window. Thought first it's like leak, but nope, don't think so.

Hmm... given lighting angle, I'm not sure I buy that arrangement producing such a sharp stripe there. Still, it's a fairly minor detail.

lw_blender
02-22-2018, 06:30 PM
41m 42s
(by the way; if you compare the render times with the ones from lw_blender, keep in mind he is running 48 threads @ 3.5 GHz while I use 32 threads @ 3.1 GHz)

140263

Very cool test, thanks for doing them! :thumbsup:

Although I'm still uncomfortable with the almost NO color bleeding in your LW 2018 renders, as if there was NO GI?
I know it isn't the case, you already replied to that question, but it still looks a bit too dull in my view.

And I much prefer the version with the walls and doors very rough, looks a lot more realistic for that type of apartment. :)

As for the render times, dont forget that my 33'21" render was done on a single i7/6700K, my Windows 10 machine, which is more in the line of yours rendered the same image in 12'41" in Kray3 RC5! ;)
Anyway, that's good to see good archviz renders done with LW2018, could be a quite good start for the future, especially if you can avoid noise in blurry reflections, as shown in your renders, without render times going through the roof! :thumbsup:

madno
02-22-2018, 06:48 PM
Oh, sorry, thought you used the dual xeon. I'll try to edit my previous post.

And you are right regarding the bleed. I have checked if there is a setting but did not find one. There should be at least some bleed from the floor. Even more so that it is reflected in the walls. Don't know at the moment. If anybody has an idea I am all ears.

And @jwede I am also not sure anymore about the strange light stripe. But to late now (nearly 3 am), need sleep.

And a last one:

36m 30 seconds (the bulb is a little noisy).

140266

rustythe1
02-23-2018, 11:21 AM
its the light coming from behind the curtain from the bottom of the area light and the shadow of the pole itself, the pole sticks out 4 inches or so, so there is a fair gap, if you move the area light back a bit it will become less obvious
140282

jwiede
02-23-2018, 11:30 AM
its the light coming from behind the curtain from the bottom of the area light and the shadow of the pole itself, the pole sticks out 4 inches or so, so there is a fair gap, if you move the area light back a bit it will become less obvious
140282

The lighter stripe is caused by the shadow of the pole? Huh?

That doesn't make sense, no way a thin pole at that distance from light source caused as wide a shadow as the entire curtain, and no way secondary-bounced light caused that sharp a delineation between the two shadows. Not seeing how direct light from window could wind up in those upper corners at all. Remember, the window's slightly sunken into the wall, not flush with the wall itself. Any light source is effectively "sunken" even further behind the window itself (or should be, anyway).

No direct light is (or should be) shooting into the room along the angle you drew, able to hit that close to the corner. There's bounce off the window sill, sure, but that's indirect (and off a fairly matte white surface). I get that the curtains and pole are maybe an inch or two away from the wall, but that still doesn't explain the lighter stripe so close to the corner itself -- the angle's too shallow.

rustythe1
02-23-2018, 01:41 PM
have you opened the scene? if you move the light back and forward the stripe moves, i highlighted a dark shadow that is front of the white stripe as it is the shadow of the pole, the light from the bottom corner of the are light is going up so the shadow would be larger than the pole, and the white stripe will be wider than the gap between the curtain and the wall, if you look you can clearly see the pole comes out from the wall a fair distance,
140283

rustythe1
02-23-2018, 01:48 PM
And if you look at all the other engines you can see its a lot more visible as i would say he is putting the area light in the room to make it appear as if its much brighter as you can see the stripe goes all the way back to the corner, this is why the 2018 renders probably look darker and need more samples as the light needs less bounce being inside the room to help the GI.
140284

gar26lw
02-23-2018, 07:33 PM
id like to see an image with all renderers output side by side with times.

Qexit
02-24-2018, 04:44 AM
id like to see an image with all renderers output side by side with times....and system specs or the times are meaningless.

rustythe1
02-24-2018, 05:02 AM
especially when certain people are only out to prove 2018 is a failure, and certain people say they are not applying post, and yet their images have been resampled to 720x960 96dpi from 750x100 72dpi in photoshop to remove all the noise and jaggies, i don't think your going to get any kind of objective perspective from these threads, and as soon as you do the 2018 renders will get ripped to shreads, and like he said above, you need someone with all the renderers who is completely non objective to all of them, you cant really compare a 5 min render in octane with someone who has 6 Titans to someone who has a single i7, if i bought 6 titans in the uk it would cost me the best part of £18000 so i would expect it to render instantly, and in the uk £18000 is close to the average persons yearly salary!

Asticles
02-24-2018, 05:07 AM
Have tested almost fully adaptive, 54min in my laptop i7 with brute force.

Very happy because it is no brainer, fire and forget quality.

140292

Now I'm trying to reduce the alias on the right frame.

rustythe1
02-24-2018, 05:15 AM
what light types are you using? (i ask as i can see you have what looks like the default direction light but its still spreading around the room nicely)

Asticles
02-24-2018, 05:25 AM
140293

Two area lights on the door and window, the sun, and three spherical lights with decreasing intensity that only affect diffuse and without falloff.

rustythe1
02-24-2018, 05:29 AM
yea, i did that originally too, but it got slammed as unrealistic and a cheat,
if you put your light samples up to 16 or so instead of 1 you can drop your aa and reflection samples a lot, and it will be far quicker as the light samples have less of a hit

Asticles
02-24-2018, 05:45 AM
Everything except montercarlo is set to 1, I want to test with adaptive because at work I always do a lot of renders of a single scene, and sometimes I get ones with noise in certain areas that I didn't checked.
Thanks for the tip, but although with fully adaptive I render in more time, I ensure the quality is controlled with the aa threshold, so I get clean renders in every camera view.

rustythe1
02-24-2018, 06:00 AM
Have you read the blog gar26lw posted about http://forums.newtek.com/showthread.php?156188-awesome-blog as I'm wondering if there is some sort of misunderstanding, difference with the adaptive engine in 2018, i get the impression its more about removing noise, than it is with edges, its almost like you can throw tons of AA but it dosnt touch edges, in fact i brought this up in another thread where if you render just edges, the aa dosnt touch it at all, if i set 1 or 200 times aa the render times and results were identical, noiseless but jaggy,

Asticles
02-24-2018, 06:03 AM
Yes, by now I think you have to downscale the image, have to make more tests.
I suppose we need some sort of subsampling, but I'm not expert.

lw_blender
02-24-2018, 07:51 AM
especially when certain people are only out to prove 2018 is a failure, and certain people say they are not applying post, and yet their images have been resampled to 720x960 96dpi from 750x100 72dpi in photoshop to remove all the noise and jaggies, i don't think your going to get any kind of objective perspective from these threads, and as soon as you do the 2018 renders will get ripped to shreads, and like he said above, you need someone with all the renderers who is completely non objective to all of them, you cant really compare a 5 min render in octane with someone who has 6 Titans to someone who has a single i7, if i bought 6 titans in the uk it would cost me the best part of £18000 so i would expect it to render instantly, and in the uk £18000 is close to the average persons yearly salary!
What make you think it HAS to be that people want to prove LW 2018 is a failure?
I know I would have much more important things to do in my life than just post bad things for a soft I don't use!
I think this thread, for the majority of the posts, is more a comparison with what is available already, sometime even for LW (Kray or Octane come to mind...), to see if LW 2018 is heading in the right direction, no?
At least it gives examples and shows rooms for optimization for the next releases of LW 2018 (or whatever it's called) and that's a good thing, me think.

Asticles
02-24-2018, 08:45 AM
yea, i did that originally too, but it got slammed as unrealistic and a cheat,
if you put your light samples up to 16 or so instead of 1 you can drop your aa and reflection samples a lot, and it will be far quicker as the light samples have less of a hit

Everything is a cheat. You have to work with a fully unbiased engine to get the most realistic image. But since we can change falloff, light exclusions, interpolated gi, etc. Who can blame us to make a biased render. Too much haters here, and I'm sorry to say that, but it is true. I know some people that don't enter here because of this.

rustythe1
02-24-2018, 09:41 AM
What make you think it HAS to be that people want to prove LW 2018 is a failure?
I know I would have much more important things to do in my life than just post bad things for a soft I don't use!
I think this thread, for the majority of the posts, is more a comparison with what is available already, sometime even for LW (Kray or Octane come to mind...), to see if LW 2018 is heading in the right direction, no?
At least it gives examples and shows rooms for optimization for the next releases of LW 2018 (or whatever it's called) and that's a good thing, me think.

don't worry, it wasn't aimed at you in the slightest, but how can you compare images when some have been edited (even when they state "no post" or "raw render" when they are quite clearly not),
and yes, Asticles, i find it hard to restrain certain comments on this forum sometimes too, as many of the replies come across as hateful towards anything LW or Newtek, if you compare the comments from here to facebook there is a big difference, unfortunately the forum is a better place to organize and refer back to,

samurai_x
02-24-2018, 09:57 AM
Soft like charmin.

UnCommonGrafx
02-24-2018, 09:58 AM
Comparisons are great.

A job to destroy the competition is nonetheless a job. They even gave them names: Shills. People spend all day posting bad things about stuff they don't use because its a job.

I applaud those who have taken the scene and optimized it. I really hope some more off the fb groups take it to work with as I see it will be a valuable source of data creation.


The Kray rendering is blazingly fast. It is hoped the next iteration of 2018 is as fast. Or nearly...

Robert

samurai_x
02-24-2018, 10:08 AM
Here's a quick test.
Light location, intensity is similar but everything else, render settings, materials, sampling, etc are totally different. Aiming for the look within 10 minutes.
Lw 2018 using the new pbr materials make it easy to make it look right but clicky to setup materials. Settings are low numbers.
Lw past is easy if you have years of presets. Settings are high numbers.

An actual 5 year old I7 3770.
lw past 5min ish
lw 2018 10min ish

http://forums.newtek.com/attachment.php?attachmentid=140294&d=1519491709

rustythe1
02-24-2018, 10:17 AM
i think your light set up for the 2018 is different, the white stripe down the walls is missing from the window area light and its also missing from the reflection on the picture on the right hand side, if there is a light there it will show the reflection as pure white, a lot less noise in the 2018 render though (see table and right hand side door) and floor textures are a lot more crisp

samurai_x
02-24-2018, 10:23 AM
Light location and angles are the same. Intensity is similar(do a clay render first when comparing renderers). Settings are not exact since they're different render engines.

lw_blender
02-24-2018, 10:24 AM
Here's a quick test.
Light location, intensity is similar but everything else, render settings, materials, sampling, etc are totally different. Aiming for the look within 10 minutes.
Lw 2018 using the new pbr materials make it easy to make it look right but clicky to setup materials. Settings are low numbers.
Lw past is easy if you have years of presets. Settings are high numbers.

An actual 5 year old I7 3770.
lw past 5min ish
lw 2018 10min ish

http://forums.newtek.com/attachment.php?attachmentid=140294&d=1519491709
Thanks for your tests! :thumbsup:

The one on the left is so much better it isn't even funny. The one on the right looks like a NO GI render, with almost no color bleeding at all, yje floor and the table are probably "a bit" too highly reflective and the wood table looks dull, not to mention the trash whic looks like gum?

Don't know which is which but the left render looks really cool for a 5 or 10 (depending on the version of LW...) minutes render, well done! :)

rustythe1
02-24-2018, 10:35 AM
Light location and angles are the same. Intensity is similar(do a clay render first when comparing renderers). Settings are not exact since they're different render engines.

well something is wrong with the set up, if you look at absolutely every other render, even in all the other render engines, and all the 2018 other renders lights are always visible in reflections, but they are missing here as if the scene is purely lit by environment, its not a negative comment by the way, just an observation that the something is a miss,
140295

samurai_x
02-24-2018, 10:39 AM
Thanks for your tests! :thumbsup:

The one on the left is so much better it isn't even funny. The one on the right looks like a NO GI render, with almost no color bleeding at all, yje floor and the table are probably "a bit" too highly reflective and the wood table looks dull, not to mention the trash whic looks like gum?

Don't know which is which but the left render looks really cool for a 5 or 10 (depending on the version of LW...) minutes render, well done! :)

Lw past is lw 11 on the right :D

Actually for me there's more bleeding on the right render if you look at the walls they're reddish probably from the bounced floor.
Lw 2018 walls look cooler tone but there's no blue in the scene.

The reflection on the table i forgot to add a blurry material. Probably add 30secs render time max.
The black trash I didn't touch for lw past. I just focused on the wall, floor, glass materials for this quick test.
Floor reflection is really dependent on taste. I wanted to make it more blurry but it adds atleast 30% more render time for both renderers.

One thing that makes lw 2018 look correct is if you switch to pbr materials they're automatically physically correct materials. No need for manual fresnel.

samurai_x
02-24-2018, 10:43 AM
well something is wrong with the set up, if you look at absolutely every other render, even in all the other render engines, and all the 2018 other renders lights are always visible in reflections, but they are missing here as if the scene is purely lit by environment, its not a negative comment by the way, just an observation that the something is a miss,
140295

That has nothing to do with the light. I changed the glass material into dielectric for lw 11 and that picture underneath showed up. If I used a standard material with a gradient reflection it will look similar to lw 2018.
If you notice the jars on the table look different but they're both dielectric with no settings changed.

djwaterman
02-24-2018, 11:55 AM
Here's a quick test.
Light location, intensity is similar but everything else, render settings, materials, sampling, etc are totally different. Aiming for the look within 10 minutes.
Lw 2018 using the new pbr materials make it easy to make it look right but clicky to setup materials. Settings are low numbers.
Lw past is easy if you have years of presets. Settings are high numbers.

An actual 5 year old I7 3770.
lw past 5min ish
lw 2018 10min ish



SamuraiX, can you describe the light set up you used and your general GI settings?

Nicolas Jordan
02-24-2018, 12:45 PM
Here's a quick test.
Light location, intensity is similar but everything else, render settings, materials, sampling, etc are totally different. Aiming for the look within 10 minutes.
Lw 2018 using the new pbr materials make it easy to make it look right but clicky to setup materials. Settings are low numbers.
Lw past is easy if you have years of presets. Settings are high numbers.

An actual 5 year old I7 3770.
lw past 5min ish
lw 2018 10min ish

http://forums.newtek.com/attachment.php?attachmentid=140294&d=1519491709

Nice test! I still haven't found a good reason to use Lightwave 2018 over 2015 in production especially if the same renderings are taking longer to achieve the same level of quality. There simply is no advantage to using it unless you absolutely need to use one of the new features for a project.

Rayek
02-24-2018, 12:57 PM
Notice the moire pattern in the table once again in the LW2018 render - which occurred during my tests as well. Never seen that happen in any render engine before: almost as if the image was printed and scanned in.

What is the cause? How can it be prevented? Is it a bug, or a specific combo of settings that causes it?

Anti-aliasing seems much better in the 2015 render. Check out the ceiling panels.

rustythe1
02-24-2018, 01:17 PM
Notice the moire pattern in the table once again in the LW2018 render - which occurred during my tests as well. Never seen that happen in any render engine before: almost as if the image was printed and scanned in.

What is the cause? How can it be prevented? Is it a bug, or a specific combo of settings that causes it?

Anti-aliasing seems much better in the 2015 render. Check out the ceiling panels.

think you got them muddled up, he said old lightwave is on the right, so that one has the error and AA is better in 2018

prometheus
02-24-2018, 02:47 PM
To me the right image look much better overall, like the tree table shadows and surfacing looks better, and overall crisper and cleaner, while I would prefer the left floor brithness effect on the floor.

madno
02-24-2018, 06:10 PM
I think samurai_x used interpolated. The 2018 shows it very obviously. The light on the walls is wobbly, uneven etc. It seems to me, the interpolated engine in 2018 has room for improvement. Regarding the reflection in the right painting. As I undestand it, lights in LW 2018 are now reflecting in spec. (which I think is a big adavantage over previous versions). Because most of the test renders are faking the way to white look (totally unrealistic with espect to the small window), the light value is high. And this shows in the reflections. I think, they are white because the light is so bright. A lot of the renders therefore also showed the blown out floor near the window.

Regarding the bleed:
I checked. It is all ok with it. The reason why it is not so obvious in the renders is, because the floor does not provide enough energy of light and saturation of color if no "sun" hits it directly. I watched rooms in real life and they look like that as well. In previous LW versions we all complained about the much to strong color bleed in GI and used node tricks to compensate. Now there is a demand for the opposite it seems :)

140300

140299

140298

140297

140302

140296

140301

140303

Renders with / without reflection on walls and with / without sun

samurai_x
02-25-2018, 12:01 AM
SamuraiX, can you describe the light set up you used and your general GI settings?

Lw 2018 low numbers
mc interp
diff bounce no more than 4 since it doesn't seem to affect the scene more than that
primary rays 64-256 is enough to rid of splotches
secondary rays 4 not much effect on this scene

cam
min/max 1/16
as 0.1
filter radius 0.5 can do 0.1 for sharp details but adds significant render time

lights
area in window, portal right door, simulated sun with area light. Vray has some tips to render Scandinavian interiors.




Nice test! I still haven't found a good reason to use Lightwave 2018 over 2015 in production especially if the same renderings are taking longer to achieve the same level of quality. There simply is no advantage to using it unless you absolutely need to use one of the new features for a project.

For new users lw 2018 will be much easier to render pretty pictures.
The materials and lights mimic real life properties like how Vray was so easy to get photoreal images 15 years ago. From the get go using PBDSF and the lights in lw 2018 gives instant good results, just not as fast as vray.

For a seasoned lw veteran with years of preset scenes and materials, hard to say. You will have to move forward at some point.


Re the lw 11 glass.
I turned off dielectric for the glass frame. Looks similar now with lw 2018. The glass jar is still dielectric but my preset scene had a sky that made it blue previously. Turned that off now its similar to lw 2018. Made the table blurry like lw 2018.
http://forums.newtek.com/attachment.php?attachmentid=140305&d=1519541877

samurai_x
02-25-2018, 12:27 AM
Turning up blurry reflections from previous post

No significant render time added for lw 2018.

Quick edits on some materials in lw 11 to make them look similar to lw 2018.
Added a lumi poly outside the window to simulate the white sheen on the floor like lw 2018.
I cranked up the aa, as and switch to mitchel sharp but still the goal is 10 mins or less.

five year old 4 core I73770
lw 2018 10min10sec
lw 11 9min40sec

http://forums.newtek.com/attachment.php?attachmentid=140306&d=1519543430

Exception
02-25-2018, 01:16 AM
Great to see Samurai,

I'm a bit behind testing this, just fired this scene up, changed the light setup to an accurate sun, skylight with portals, and changed all camera, light and render settings to my known best quality ones for a scene like this. I like the direct sunlight, but I'll remove it for testing purposes in the next iterations. Also, I see caustics was on which probably caused that light splotch top right.

Not very useful to compare, but I think it's nice to see that LW2018 can crank out a reasonable quality picture, if you have patience :D (>8 hours eek).

Do not like the blotching, the AA could be better, and there's other details that are bothersome, I hope to address when optimizing.

http://forums.newtek.com/attachment.php?attachmentid=140307&d=1519546587

140307

samurai_x
02-25-2018, 07:03 AM
8 hours?? Maybe really high settings.


Further testing. I'm encountering fireflies now when the scene has stronger lights and darks.

8min20sec
http://forums.newtek.com/attachment.php?attachmentid=140311&d=1519567325

gar26lw
02-25-2018, 07:34 AM
shame about the fireflies bug :/ when is 2018.03 out?

samurai_x
02-25-2018, 07:42 AM
https://docs.lightwave3d.com/display/LW2018/Removing+Noise+workflow

djwaterman
02-25-2018, 02:46 PM
I messed around with the scene over the weekend, but with the sort of light set ups you described, I could never get enough light into the room to make it bright like yours, unless I added another light somewhere inside the room. I also tried an area light for a sun but I didn't know how to set that up to get the directional lighting through the window that a sun light achieves. I'm sort of mystified to see these nice bright renders, how are you getting them?


Although having just loaded up the original LW scene I see that he had his area lights with no fall off, so that might be it.

rustythe1
02-25-2018, 05:40 PM
its because most people are turning the light falloff off and using linear, it clicked for me earlier as i was getting nice renders but the light was much brighter at the window end, i turned the lights to linear and they suddenly looked the same, in reality the renders should look more like these, and yes there are some errors in these as that's not what I'm uploading them for, but the second one is more how it should look in linear CS, the first one is just increasing the light multiplier and setting to srgb,
140320140321

gar26lw
02-25-2018, 07:43 PM
so keep lights on linear cs at all times?

madno
02-26-2018, 12:16 AM
No, only if you intend to get the "all bright" room look.

What I undestand from testing:

BSDF + lights with fall off acts like real world - or better like photographs.
Light outside the room has only small window to enter -> room is bright in the area of the window frame but gets quickly darker in the inside.
Like with a photo. Either the outside is exposed correctly (visible details instead of blown out) or the room inside is exposed correctly but still with darkening happening due to fall off.
But, as we want a bright look overall, we switch the falloff to linear. And in most cases even put a fill light somewhere in the room (e.g. behind the camera).
Other option might be to output exr / hdr (= enough detail data inside and outside the room) and adjust everything in post (have not yet tried this).

samurai_x
02-26-2018, 12:47 AM
My light rig uses inverse. That linear light looks overblown.
Hopefully there will be a physical camera in lightwave to make it easier like vray.


This render uses a different light rig with a directional light for moody pics.

I73770
Previous 8min, now 1hr24min to get rid of spec noise and fireflies but still noise on the table. No noise reduction cheat. Too long render time imo.

http://forums.newtek.com/attachment.php?attachmentid=140324&d=1519631050

Exception
02-26-2018, 12:52 AM
Down to about 2.5 hours render time at full res... some parts look good but the inset part on the left is horribly splotchy.
AA looks pretty nice here at 10/32. Except for the excusable high energy blowouts on the glass which need HDR clamping.

http://forums.newtek.com/attachment.php?attachmentid=140325&d=1519631492&thumb=0&stc=1

140325

samurai_x
02-26-2018, 12:59 AM
Is that lw_blender's scene?
Your first render with the sun doesn't have fireflies but rendered 8 hours? :D
I have to check his scene if the table material has something that removed noise and fireflies.

madno
02-26-2018, 01:59 AM
I tried with interpolated at insame values - up to 8000+ primary samples on the door and room object. Even with those values the splotchies remain in the left door tunnel. It seems to me, they can only be removed with an extra light filling that area. Or skip GI and cheat it like rustythe1 has shown it (he got 80 something seconds that way). Or brute force.

140330

24m58s (1500 x 1126 | dual xeon @ 3.1 GHz | 32 threads)


GI BF
Slight LW noise reduction (but still noise there)
Fill area light behind the camera (affect only diffuse and shadow, no spec.)
Fill area light in the open door on the right (this uses inverse, all other are linear)
Main area light in front of window (moved it a bit back to get rid of the light streaks on the wall left and right form the curtain)
No portals at all (in testing they took time with no advantage - there is no background to sample relevant information from)
Floor planks have a color tool with a little saturation and contrast boost applied (maybe a bit to much)
Reduced blackpoint to 80% in LW render editor to get a bit more contrast (but real styling I would do in post)
Background image with less luminosity (according to the composition it is in shadow)
Unsure about the vase - should'nt it look more clear? Used 4 refraction recursions


Advantage of this render is, that even the jpg has headroom for post. Nothing is blown out etc.

But, I still can't get a ready made final render without the need to post out in a few minutes. But for me 24 minutes is ok. I would even give it more to eliminate the noise (but then I would add new noise (bump or so) to get rid of the way too flat look of the walls).

Material setting are from the table.
140333 140334 140331 140332

Forgot, I sub-ded the table top and the lamp to help the AA on the harsh edge highlights.

samurai_x
02-26-2018, 02:40 AM
24m58s (1500 x 1126 | dual xeon @ 3.1 GHz | 32 threads)


24m58s on a dual xeon @ 3.1 GHz probably 2.5 hours for my I73770. :D
That's extremely long for a simple scene like this.

Exception
02-26-2018, 04:12 AM
Is that lw_blender's scene?
Your first render with the sun doesn't have fireflies but rendered 8 hours? :D
I have to check his scene if the table material has something that removed noise and fireflies.

Yes, but I changed almost everything except the materials.
The first pass I didn't optimize anything, just set it at something high and let it go. I think the fireflies are just too few aa samples in the blurred reflections.

MichaelT
02-26-2018, 06:24 AM
There's something odd going on with the curtain shadows against the upper corners of the back (window) wall in that one. See the odd light vertical stripe present in both sides' curtain shadows?

That is reflected light off the back of the curtains, you should see that unless the curtains are on the actual wall directly. Is a stripe because the walls off the side where the light is coming from casts a shadow of their own.

MichaelT
02-26-2018, 06:26 AM
shame about the fireflies bug :/ when is 2018.03 out?

Fireflies are not a bug. They exists in any renderer that calculates light like this. Our job is to find the best way to deal with them. With this type of renderer the whole image is based on noise, it is the very reason we see an image at all.

samurai_x
02-26-2018, 06:36 AM
Yes, but I changed almost everything except the materials.
The first pass I didn't optimize anything, just set it at something high and let it go. I think the fireflies are just too few aa samples in the blurred reflections.

I tried to render that scene and thhe first few minutes I can only see a few squares on the render. :D
My pc can't handle it.
What pc did you render to get 8 hours?

RPSchmidt
02-26-2018, 06:45 AM
I tried with interpolated at insame values - up to 8000+ primary samples on the door and room object. Even with those values the splotchies remain in the left door tunnel. It seems to me, they can only be removed with an extra light filling that area. Or skip GI and cheat it like rustythe1 has shown it (he got 80 something seconds that way). Or brute force.

I noticed you didn't use mipmapping; is that just a bozo no no for archvis, or a personal choice?

I would think that mipmapping would help smooth out splotches a bit by adding some artificial anisotropic, especially at the angle tolerance you are using on global illumination, but I could be wrong.

samurai_x
02-26-2018, 07:18 AM
4 core I73770

lw 11 8min
lw 2018 51 mins. Down from previous post of 1hr24min. Specular indirect noise(GAWD awful in this renderer) on the table is very stubborn. Higher reflect samples, AA samples "helped" a bit. AS sampling doesn't do anything in this renderer afaict.

http://forums.newtek.com/attachment.php?attachmentid=140345&d=1519654491

ianr
02-26-2018, 08:25 AM
Fireflies are not a bug. They exists in any renderer that calculates light like this. Our job is to find the best way to deal with them. With this type of renderer the whole image is based on noise, it is the very reason we see an image at all.


They should by now (Devs) looking at an semi intelligent

Ai de-Noiser (an auto-runner) like Octane are into

And also studying RedShifts tuneable approach

But who is talking that back to us over this ?

JamesCurtis
02-26-2018, 09:35 AM
Is there a LW 2018 scene file that I can use to do these tests? I was looking through the posts, and all I found was one that rendered everything in grey tones.

Exception
02-26-2018, 01:44 PM
Is there a LW 2018 scene file that I can use to do these tests? I was looking through the posts, and all I found was one that rendered everything in grey tones.

LW_Blender posted this download link with a 2018 version: https://d.pr/f/UCgXlA

Exception
02-26-2018, 01:46 PM
Here's a sample analysis for those interested, at 5px min 200 px max and 40 degrees tolerance.
See the massive amount of samples on the curved surfaces. the degree tolerance should reduce these, but that doesn't actually seem to work... maybe there's room for optimization improvement there for sample placement.

http://forums.newtek.com/attachment.php?attachmentid=140359&d=1519677974

140359