Show me that Lightwave 2018's new render engine is a worthy successor to 2015...

[Paul_Boland]

Hi Folks.

There was a recent heated debate on the Lightwave Wiki Facebook page about the render engine in Lightwave 2018 vs Lightwave 2015. A moderator of that page has closed that thread. There is a lot of split opinions about the Lightwave 2018 renderer and if it's a worthy successor to 2015? After playing around with it for over two months, reading the online documentation from cover to cover, and watching many videos on YouTube about it, as of this point I'm not convinced. Every test I do in Lightwave 2018 is a hell of a lot slower than the same test in Lightwave 2015 and the render results are not as good. And I'm not a lone voice in the wilderness saying this.

Whenever this debate comes up, all I keep seeing from the pro-Lightwave 2018 group is "It's a new engine, you have to learn it's new ways." Well, folks, I've tried... I've tried damn hard to learn it. And all I see is an inferior engine in 2018. I decided to be fair and to try and pit Lightwave 2018 and Lightwave 2015 against each other on even terms, creating the same scene file in both packages, sorting out the render setting so they match, and texturing them the same too, and the results were jaw-dropping bad for Lightwave 2018. You can view the video on this here:



The end result...


Lightwave 2015 rendered this in 46.8 seconds.


Lightwave 2018 rendered this in 9 minutes 48 seconds!!

So the reason for this topic I'm posting here is, I want someone to show me, help me, to see that the Lightwave 2018 render engine is a worthy successor to Lightwave 2015! Tell me what I need to do shave 9 minutes off the 2018 render time, bringing it in par with 2015, producing a crisp, clean, beautiful render on par with Lightwave 2015 in around 46 odd seconds. I'm attaching both scene files below in the zip file. Show me where I'm going wrong, prove to me that Lightwave 2018 is a worthy successor, and I'll create a follow-up video highlighting the facts. Because right now, the render times and end results in Lightwave 2018 compared to Lightwave 2015 leave a lot to be desired...

[mummyman]

Oh.. boy.. here we go. My day is planned!

[Chuck]

Whenever this debate comes up, all I keep seeing from the pro-Lightwave 2018 group is "It's a new engine, you have to learn it's new ways."

That's not all that's been said and I've seen a number of posts from RebelHill, Elusive Elephant, Matt Gorner, Antti Järvelä and others who have been working with 2018 for some time that they can get equal or better quality with lower rendering times for most cases, and they've discussed the settings that need to be experimented with for those results. One of the things I have on my task list is to gather up the tips, pointers and commentary on use of the new renderer into some blog entries, which I hope will help folks along the learning curve.

That said, I certainly look forward to those folks and others taking a look at your scene and working with it. It may also turn out to be of interest for development, because as a new renderer it is likely that there is plenty of room for reviewing use cases and tuning for better performance. It is a good idea, when you have a scene that seems to be intractable in getting to a relatively equal or better time to 2018, to submit that as a bug report.

[jeric_synergy]

That's not all that's been said and I've seen a number of posts from RenderHill, Elusive Elephant, Matt Gorner, Antti Järvelä and others who have been working with 2018 for some time that they can get equal or better quality with lower rendering times for most cases, and they've discussed the settings that need to be experimented with for those results. One of the things I have on my task list is to gather up the tips, pointers and commentary on use of the new renderer into some blog entries, which I hope will help folks along the learning curve.

While such is aggregation is GOOD, what would be even better would be if such articles were linked IN THE DOCUMENTATION.

I've been encouraging this for years, and that it would be totally under LWG's control (versus worthy but independent online articles) makes it even more of A Good Idea.

There's lots of worthy material out there that would enhance the LW documentation: I don't see any point in not using it, even if some fraction of it is ephemeral. To not have it in the dox just makes life harder for LW users.

[darkChief]

Hi Folks.

There was a recent heated debate on the Lightwave Wiki Facebook page about the render engine in Lightwave 2018 vs Lightwave 2015. A moderator of that page has closed that thread. There is a lot of split opinions about the Lightwave 2018 renderer and if it's a worthy successor to 2015? After playing around with it for over two months, reading the online documentation from cover to cover, and watching many videos on YouTube about it, as of this point I'm not convinced. Every test I do in Lightwave 2018 is a hell of a lot slower than the same test in Lightwave 2015 and the render results are not as good. And I'm not a lone voice in the wilderness saying this.

Whenever this debate comes up, all I keep seeing from the pro-Lightwave 2018 group is "It's a new engine, you have to learn it's new ways." Well, folks, I've tried... I've tried damn hard to learn it. And all I see is an inferior engine in 2018. I decided to be fair and to try and pit Lightwave 2018 and Lightwave 2015 against each other on even terms, creating the same scene file in both packages, sorting out the render setting so they match, and texturing them the same too, and the results were jaw-dropping bad for Lightwave 2018. You can view the video on this here:



The end result...


Lightwave 2015 rendered this in 46.8 seconds.


Lightwave 2018 rendered this in 9 minutes 48 seconds!!

So the reason for this topic I'm posting here is, I want someone to show me, help me, to see that the Lightwave 2018 render engine is a worthy successor to Lightwave 2015! Tell me what I need to do shave 9 minutes off the 2018 render time, bringing it in par with 2015, producing a crisp, clean, beautiful render on par with Lightwave 2015 in around 46 odd seconds. I'm attaching both scene files below in the zip file. Show me where I'm going wrong, prove to me that Lightwave 2018 is a worthy successor, and I'll create a follow-up video highlighting the facts. Because right now, the render times and end results in Lightwave 2018 compared to Lightwave 2015 leave a lot to be desired...

Don't know about matching speeds, but I improved the times on the scene you shared. I switched on interpolated Gi (primary rays 32, secondary 16. Double each value if you notice splotches). Lowered the camera as samples (minimum 1, maximum 4). And reflection samples between 4-8 gives good results with noise filtering.

Also Principled BSDF takes longer than Standard materials.

Also Gi doesn't seem to make a big difference for the scene, switching it off also improves times.

Hope that Helps in some way.

[prometheus]

I think some things look better, was testing some daz figures and using the new skin material vs old looks very nice in 2018, and alos eye reflection with dielectric etc looks very nice..and you get what you see in vpr ..propagating through to final render.
Some landscape terrain setting ..in general principled look better, though I still have to discover the Additive sharpness..and see what in lightwave 2018 that replaces that setting from 2015, addivite sharpness for rocks was great, but I do not see a way for the new shaders to work with that..or I have just missed it, it may not be necessary.

What bothers me is VPR speed refinement together with GI, where I feel it is quite a bit faster in 2015 (VPR refinement) different beast though...and despite me lowering as many values I can for rendering, it still iterates slower than the old GI in 2015, when it comes to final renders..that is a different matter that I haven´t looked in to that much.

The new volumetrics, at lower scale they look great when considering render speed, at larger real life scale of clouds, the balance between render speed and step quality isn´t acceptable, in fact the old legacy hypervoxels(within 2018) is blistering fast compared to the new volumetrics.
the old legacy hypervoxels have removed a barrier ( probably the new render changes, not hv changes) so you can overlap hypervoxels in a way with several nulls which you couldn´t do in 2015 without extreme lag.

[Kryslin]

Paul:

While I can't match the 48 second render time, even I, rank amateur that I am, can get as good as the 2015 render in less than 2 minutes.

The key is samples (Light, Radiosity Rays, Reflection, Refraction, Subsurface Scattering) over AA passes, in nearly all cases. Also, don't use Adaptive AA. Set your number of passes and let it go (Adaptive AA increases render times by 10-15% in my recent experience). You start at 1 pass, and samples set at defaults, and then bring up the sample values where the noise is. Then, start bringing the AA passes up. The noise reduction wortkflow in the documentation does work well. There are exceptions to the samples before passes rule; High Frequency textures and Fiber FX are the two that come to mind.

[kolby]

Here is my test.
LW 2015.3 - 26,5 sec.


LW 2018.0.2 - 25,6 sec.


Only disaster are fireflies. And I can't match that crazy transparent material on right sphere.

[wesleycorgi]

Hi Kolby: can you share what you tweaked to get down to 25.6?

[Morgan Nilsson]

Oh boy, there is so much wrong with the settings in the scene its no wonder you think the LW2018 renderer is crap...

First of all, camera samples, 10/20 adaptive AA with a threshold of 0.01? That is pure insanity. I never go above 8 as a maximum (and that is in scenes with an insane amount of edges that can cause stair-stepping artifacts, those settings has nothing to do with noise for me) and rarely needs to change minimum from 1.

Threshold I typically go around 0.05-0.08, but like all settings, it varies from a scene to scene basis.

Also, you get a ton of indirect diffuse noise with gi rays set to 1 only, around 6 is a decent value, and, if you have specific objects that suffer a lot from indirect diffuse noise, you can just go into that objects properties and force more gi rays on it to resolve the noise.

Camera samples should never, ever be the go to setting to remove noise in your scenes. Those should in 99/100 cases only be tweaked to fix issues with sharp edges that causes the stair-stepping artifacts.

[rustythe1]

Yep, I changed camera, surfaces, (turn off ray back, glossy etc) turn off GI as its a point light in a closed box, GI has no effect, changed a few other settings and got it down to 3 seconds

[kolby]

Hi Kolby: can you share what you tweaked to get down to 25.6?

updated contents
2015vs2018_kolby.zip

[wesleycorgi]

Thanks!

[SBowie]

Paul, my friend, I want to say I feel your pain - but instead I have to say that I only look forward to feeling it. I haven't had time yet to start mounting the learning 2018 curve. I'll do so with a personal project sometime after NAB. Thankfully, I have an advantage, in that I can walk over and ask Jarrod and Matt stupid questions from time to time (since their desks are within 20 feet of mine.

This said - my impression, supported by a conversation we had at lunchtime, is that 2018 really does represent a quantum leap, albeit one that comes with a corresponding (one-time) learning curve. And I think helping people surmount that curve represents a real challenge for both NewTek and many of its longstanding LW customers (including you and me). What I've been confidently told is that some renders may indeed take a little longer, though not always - but that the payback in terms of render quality, time saved in surfacing setup, etc., more than make up for this. To paraphrase Jarrod, 'Once you get the hang of the PBR renderer, you will never consider going back'.

I'm happy you went to the trouble of posting this thread, and am sure the feedback will go a long way to helping me, and others.

[Bitboy]

Paul, I've been doing a little interior scene and I'm pretty happy with LW 2018 so far. Still learning the new stuff, but here's my image which took ~20 minutes to render on an older i5 4430S @ 2.70 Ghz with 12 GB RAM. And I'm not very well versed in doing interiors at all, so for a first try in LW 2018 this is ok in my book. I'll probably adjust some of my settings with the advice people have been writing in this thread, so thanks for that guys

And even with that rendertime, I'll probably be able to bring it down a bit, since it's not fully optimized yet. I know there's a little noise in the image a few places and there's still stuff missing that I have to put in. All surfaces are Principled, no Standard materials at all. I've attached my settings and they're not that different than the standard settings, but I've adjusted the GI and Render settings some to get the quality where I wanted it.

So I don't get the complaints since I never had my interiors looking like this in 2015 with a decent render time, and that was on a faster machine as well. Just my 2 cents...

Rendered image:


Render screenshot:


LW screenshot:


Camera settings:


GI settings:


Render settings: