PDA

View Full Version : Kray official announcment is coming



S0nny
04-27-2016, 01:40 AM
Seems like the website is updating.
Janusz talked about an official announcement for the beta a week ago, still don't know if it'll be public or not, but probably more info are coming this time.

Edit: yes, looks like there's something in the black background

133584

djwaterman
04-27-2016, 04:06 AM
Looks promising doesn't it?

OFF
04-27-2016, 06:59 AM
can't believe ) it's been so long..

Niko3D
04-27-2016, 07:38 AM
I'm positive, also given the latest Janusz's render on the forum...
We'll see...because the other renderers have made great strides in the meantime...

mav3rick
04-27-2016, 09:08 AM
i would say... about time :)

Snosrap
04-27-2016, 11:15 AM
I was interested - last year. :) No I'm waiting to see what NT has to offer with LW Next. Then I'll re-evaluate. Crappy timing for Kray.

S0nny
04-27-2016, 02:55 PM
There are very few information about both Next and K3 for now, and some simple scenes with just volumetrics or sss aren't really telling me nothing about the new lw render engine capabilities.
For K3, the lack of updates for years were alarming to say the least, but now a new dedicated web site is coming, hopefully with some more info and maybe a beta to test.

I don't know, I really just hope a new glorious future for lw, let's see what happens.

hrgiger
04-27-2016, 03:00 PM
Autodesk probably bought them.

S0nny
04-27-2016, 03:40 PM
Really! They need to stop to phagocytize everything, this is a meme already ;D
But seriously, I think Gregor (the developer) is getting probably more money from SU Podium than kray, until it remains just a lw thing there's no such risk like autodesk.

erikals
04-27-2016, 09:48 PM
http://www.kraytracing.com

...so, wonder how this will stand up to the new LightWave 2016 Render Engine...
...and Octane

ideart
04-28-2016, 02:15 AM
http://www.kraytracing.com

...so, wonder how this will stand up to the new LightWave 2016 Render Engine...
...and Octane

It depends I guess on what do you want to render.
Lightwave's render engine has no restrictions like the other renderers and is good enough for many tasks except productive archiviz.
Kray is the best, fastest and most productive in archiviz stills but with a couple of bugs in animation.
Octane is great and we get access to more advanced materials and I managed to to archiviz but then again I have 8 980's for rendering so it is quite expensive solution and with it's own limitations.
Personally I am waiting for Kray 3 to see if it would be enough to keep me in Lightwave. Lightwave is very neglected by 3rd party renderers.

S0nny
04-29-2016, 06:23 AM
So, website up and running, but not much info about k3 for now. There's a live test in java and the announcement about beta available for developers... but I'm not a developer, so I can't test much for now.
Anyway I think it's still in progress, my login credential doesn't work and some part of the website are not completed yet.

edit: notice that the sdk is already available for other apps to integrate kray.

lightscape
04-30-2016, 09:40 PM
http://betablox.com/wp-content/uploads/2015/10/Kansas-City-Missouri-Entrepreneur.jpg

Has it been 3 years on life support? Can't recall anymore.

erikals
04-30-2016, 11:41 PM
Kray 3 is a rewrite as far as i recall, hence the looong wait... http://erikalstad.com/backup/misc.php_files/smile.gif

lightscape
05-01-2016, 05:18 AM
G really needs to diversify his market. Sketchup was profitable. But to focus on lw and sketchup alone might not be enough to stay afloat and hire another coder to speed up development.

Ah well its too late anyway. There's no other renderer that can beat Redshift now.

erikals
05-01-2016, 05:37 AM
Redshift seems a bit hyped, doesn't it?

some notes,
https://render.otoy.com/forum/viewtopic.php?f=9&t=53042

+ we haven't seen the new LightWave Render Engine speed, it might be something cool.

i don't know, i just like the Octane quality, it does produce noise, but much can be removed in post,


but back to RedShift, 1min render per frame for this quality is pretty good.
https://www.youtube.com/watch?v=GqRR-_oV7_E

or 3min if one adds Dof

THIBAULT
05-01-2016, 09:30 AM
Thea render is good too ! CPU + GPU interesting

erikals
05-01-2016, 09:33 AM
do note, Octane will soon have CPU + GPU

S0nny
05-01-2016, 10:08 AM
There's no other renderer that can beat Redshift now.

Interesting, can you elaborate more on that?

lightscape
05-01-2016, 10:18 AM
Redshift seems a bit hyped, doesn't it?


http://eu.blizzard.com/en-gb/company/careers/posting.html?id=160009R

When a studio like Blizzard adapts it in their pipeline you know its proven. ;)
Listed beside RenderMan, Vray, its in good company.

erikals
05-01-2016, 01:52 PM
yeah, there are so many render engines now though, hard to keep track.

and one year Render Engine X gets the most attention, the next year it's Render Engine Y.

then finally Render Engine Z gets a rewrite.

maybe one decides to stick with Engine Z after all, as it has free render nodes and renders CelShading beautifully.

lightscape
05-02-2016, 08:06 PM
yeah, there are so many render engines now though, hard to keep track.

and one year Render Engine X gets the most attention, the next year it's Render Engine Y.

then finally Render Engine Z gets a rewrite.

maybe one decides to stick with Engine Z after all, as it has free render nodes and renders CelShading beautifully.

For me I have two criterias to go with a renderer. Its either the most used or the cheapest renderer.
Lightwave is cheap with 999 rendernodes. Its a keeper.
Vray is the most used, industry standard and you won't go wrong knowing it.

The next renderer that will dominate soon imo would be Redshift.

THIBAULT
05-02-2016, 11:01 PM
For you, why Redshift and not Octane ?

S0nny
05-03-2016, 02:00 AM
Maybe it's me, but from what I see the interesting part comes from the out of core architecture in Redshift, so you don't have the polycount limit or texture size limit with low ram graphics card.
It's a gpu only render, with no dedicated plugins other than the usual Autodesk stuff (sigh).
IMHO, with this gpu only renders (like octane, after all) you are going to spend a lot for a workstation in terms of graphic cards and electric bills, but still invest a good money anyway for the cpus and ram because the other tasks requires much power anyway, so you are not going to save money on this.
I don't know about renderfarm or distributed render on gpus because I'm not an expert in this, but can be another cost.
So in the end I think that - for now - gpus unbiased renders are best suitable for animations with no super high resolution. Cpus for stills, archiviz (interiors expecially) etc.

edit: yes, Redshift is not brute force only, but I think that the argument is valid anyway.

MichaelT
05-03-2016, 03:00 PM
The page is up now it seems (Kray 2.xx) I could swear it said 3 at one point.. but maybe I was in during edit.

50one
05-03-2016, 03:34 PM
It's just around the corner™

lightscape
05-04-2016, 03:37 AM
For you, why Redshift and not Octane ?

Speed. Redshift is a biased gpu renderer. Think vray but coded for gpu. Its the fastest renderer I've used in maya.

Schwyhart
05-04-2016, 10:54 AM
Just tried the demo of Kray. I can't figure out how to apply materials. Does it use Lightwave's materials? Surely not...

jwiede
05-04-2016, 05:27 PM
Just tried the demo of Kray. I can't figure out how to apply materials. Does it use Lightwave's materials? Surely not...

If you mean the Kray plugin for LW (as in, pre-v3 demo), then yes, LW Kray actually does use LW materials.

Rumor has it if v3 ever releases, then LW Kray v3 will switch to using special Kray materials, but I suspect only G & co. know for certain.

lightscape
05-04-2016, 08:04 PM
Just tried the demo of Kray. I can't figure out how to apply materials. Does it use Lightwave's materials? Surely not...

It does.

ideart
05-05-2016, 01:46 AM
Just tried the demo of Kray. I can't figure out how to apply materials. Does it use Lightwave's materials? Surely not...

It uses Lightwave's material layers system and nodes but no shaders.
Also some nodes like carpaint, conductor and dielectric causes problems and slow render times so best not use them.
Janusz has posted in Kray forum a great set of node based materials that you can use.
Also blurred reflections are ultra fast in Kray.

Janusz Biela
09-08-2016, 01:12 AM
Hello!
here is all necessary info about K3 development:
- Kray3 has completely new code (this is no update of K2) with modular code construction
- support also upcoming new Lightwave
- support CPU with 64+ threats
- development of K3 started over 4 years ago (ask coders how long takes writing Photon Map engine...) It is million times more difficult then Path Tracing
- Support CPU/GPU in the same time (I hope soon first tests)
- Support all tone mapping in real time (white balance, burn out, exposure, blending, LWF, gamma, dark/bright Clip, contrast - some of them are not ready yet)
- Support all system renders in real time - include: the first on the world Full PhotonMap system (details what is this, on the Kray forum)
- The fastest on the World Photon Mapping system
- K3 physical light system include LEM lights with direct light behavior
- The newest Physical light system with Hosek, now working also in PhotonMap system
- Auto RAM GI store and Auto Update GI in Kray3VPR
- soon also KrayShaderSurface system with his own physical surfaces (We need still LW Nodes because Lightwave has the best pre processing Node system but Shaders must be done inside K3Core - this is necessary for fast render without errors)

This is only what happen now. So if you want update information please visit Kray web forum.
I update non stop develop progress...so almost everyday news :)

Janusz Biela
09-08-2016, 02:04 AM
usually I do not post on Newtek Forum but here is small exception because these two Kray3VPR interior renders are first renders in Full PhotonMap system in whole Kray history:
With 200.000.000 photons positioned on the whole scene surface (~80 sec).

Kray3VPR QMC (uncache) 2 min with 10 bounce lights:
134389

Kray3VPR 40min pure Path Tracing 10 bounce lights:
134386

Differences in renders are from my bug haunting fight (sometimes scenes are different) but at least we can see approximately time renders.

here is standard render with QMC (PhotonMap+QMC)
134388

erikals
09-08-2016, 02:31 AM
nice, but it needs AA

Janusz Biela
09-08-2016, 04:05 AM
nice, but it needs AA

Not in this phase - is not necessary.
You can 30-70% more time to get complete time render.
AA is not activated yet in Kray3VPR because G. wants add new one in K3.1.

samurai_x
09-08-2016, 04:28 AM
Why is there a huge difference in render time between the two?
Pathtracing render has floating models. :D

Janusz Biela
09-08-2016, 08:38 AM
Why is there a huge difference in render time between the two?
Pathtracing render has floating models. :D

The quality of GI is exactly the same (only professionals can see a slight differences)
Differences in time because one is Path Tracing (try Maxwell, Octane) where is nothing interpolated ...in other word: must be slow, it is slow and will be slow always.
Faster render is done by Photon Mapping in PhotonMap system where Global Illumination sampling is calculate by Quasi Monte Carlo system.
Sampling of QMC is very similar to sampling of Path Tracing (there is some differences especially in MTL system but I can not render in MTL because we do not have yet this in K3)
The difference is in first phase where the base for calculating GI is Photon Mapping.

PM+QMC (fast render example) is what you have in Corona render but in my opinion way faster.

Floating object - of course! because missing furniture and object in scene (for haunting bugs purposes) :)
There is no problems with contact shadow, light leaks and splotches in this system. We have just quality of Path Tracing with speed render x10-20 .

Janusz Biela
09-10-2016, 02:20 AM
I made small test K3VPR - LW VPR with the same conditions both 10 bounce lights with QMC sampling (also some differences in scene: LW has no Physky so test has no compare speed purposes):

Lightwave VPR
after 2 sec:
134415

after 40 sec:
134416

after 3 min:
134414

Kray3VPR
after 2 sec:
134417

after 20-50 sec (missing only Anti Aliasing)
134418
134419
134421
134420

Janusz Biela
09-10-2016, 05:15 AM
Small update (I have to prepare this scene for for tutorial because maybe today will be update of K3OB3)

Kray3VPR pure Path Tracing 10 bounce lights (physky2 + Sun + Kray3AreaLights+LEM) 3min30 sec:
134428

Kray3VPR PM+QMC with 40.000.000 photons and 10 bounce lights in PhotonMap system (physky2 + Sun + Kray3AreaLights+LEM) 9 sec:
134427

jasonwestmas
09-10-2016, 06:37 AM
I`ve already checked out 4 other render engines since the last kray announcement. . . .crappy timing indeed, the rendering market is saturated. However the whole interrior lighting thing is a popular thing to perfect. Kray 3 is looking good in that regard but so was kray 2.

rustythe1
09-10-2016, 07:32 AM
what hardware are you using for these renders?

samurai_x
09-10-2016, 07:32 AM
I don't remember qmc uncached being fast. It was almost as slow as pathtracing.
Some kind of magic with K3.
Tutorial with voice please like lwcad tutorials.

Janusz Biela
09-10-2016, 09:24 AM
I`ve already checked out 4 other render engines since the last kray announcement. . . .crappy timing indeed, the rendering market is saturated. However the whole interrior lighting thing is a popular thing to perfect. Kray 3 is looking good in that regard but so was kray 2.

yes this is really annoying: full market of render engines with fancy menu, millions surfaces but with only one problem everywhere: the speed.
I just watched yesterday on Youtube Arnold Render Engine with additional stuff (surfacing, extra features, etc) really great stuff! ...but is enough read every second comment:
YOU NEED RENDER FARM!!!! This is not for 99% users....

Personally I do not understand why people believe or wants believe that Path Tracing is fast render by CPU or GPU. This is impossible. You want fast, buy more CPU (really more) or 2-4 GPU top market cards.

Janusz Biela
09-10-2016, 09:33 AM
what hardware are you using for these renders?

Depends:
- I7 CPU
- Dual Xeon 72 Threats
- Dual Xeon 32 Threats

generally around 2000-3000 pts in Cinebench R15

Janusz Biela
09-10-2016, 09:58 AM
I don't remember qmc uncached being fast. It was almost as slow as pathtracing.
Some kind of magic with K3.
Tutorial with voice please like lwcad tutorials.

Quasi Monte Carlo is very similar to the Path Tracing. The difference is in first phase.
In Path Tracing you do not have interpolation and phases. Photons and samples GI are calculated by rays/pixel in the same time.
In QMC+PM first phase is Photon Mapping (ultra fast) second phase sampling QMC.
The art is to make firsts phase (Photon Mapping) fast and accurate and actually do not exists that engine. Let see when we will close K3 Core...maybe we will be the first.
Also, very important , QMC sampling speed calculations depends strongly from information of Photon Map. In K3 Photon Map is staggering - why QMC is calculated very fast....but let`s wait for new Final Gathering in K3.1. This can be 3-6 times faster then QMC with almost same quality.


This magic is no magic: just correct and clean Photon Map make less "stress" for engine when calculate Global Illumination.
Why I shoot enormous amount of photons with small resolution. With this system I am very, very close to Path Tracing but with incredible speed render.

Unfortunately, English is not my native and I have to catch somebody who speaks naturally.

jasonwestmas
09-10-2016, 04:35 PM
yes this is really annoying: full market of render engines with fancy menu, millions surfaces but with only one problem everywhere: the speed.
I just watched yesterday on Youtube Arnold Render Engine with additional stuff (surfacing, extra features, etc) really great stuff! ...but is enough read every second comment:
YOU NEED RENDER FARM!!!! This is not for 99% users....

Personally I do not understand why people believe or wants believe that Path Tracing is fast render by CPU or GPU. This is impossible. You want fast, buy more CPU (really more) or 2-4 GPU top market cards.

yeah I'm definitely interested in biased rendering still cause it's faster. I've also had the pleasure of trying out a desktop with two video cards and redshift. One for display and the other for nothing but rendering with cuda. Nothing slows down the process, it's very fluid when tweaking scenes while you are progressively rendering.

Janusz Biela
09-11-2016, 02:00 AM
When you have two GPU top cards is not so bad. It is still slow as hell because there is no interpolation. It is really ok. But I do not like one think with GPU rendering: the limits with bounce lights. This brings artificiality in renders. Nobody today try render in Path Tracing with 10 bounce light...this is mission impossible for interiors. The amount of calculation raise up dramatically after each recurse. When you have 2-4 bounce lights is ok but when you raise up over 5 you will never finish render.
personally I am not interested in with render image in 1-5 h. I need the score in 1-5 min. It can be even less quality.
Global Illumination today is important but more important are shaders. We do not have it yet so always renders will have a bit less quality then Octane, Maxwell, Corona. There is no way also to use LW Shaders (anireflection., SSS, Dielectric and others) we will start create KrayPluginShader because physical surfaces must be render directly in Core Engine. We will keep LW pre-processing Node system - is great.

jwiede
09-11-2016, 11:08 AM
Global Illumination today is important but more important are shaders. We do not have it yet so always renders will have a bit less quality then Octane, Maxwell, Corona. There is no way also to use LW Shaders (anireflection., SSS, Dielectric and others) we will start create KrayPluginShader because physical surfaces must be render directly in Core Engine. We will keep LW pre-processing Node system - is great.

Approx. how long before such shaders exist for K3(.x)? Does that mean K3.0 won't have any material support for SSS, environmental fog, or dielectric-type transparent materials at release?

Likewise, talking about better AA in K3.1 is nice, but what are K3.0 owners supposed to do during the (likely extended) period between K3.0 and K3.1? Not all AA problems can be solved by throwing more samples, esp. where fine wires in transparency-masked fences, or above-ground distribution lines (power, phone, cable) are involved. Is there just no viable solution to such cases in K3.0?

I've been a strong proponent of Kray (2) for a long time, and I really want K3 to succeed, but am very concerned about the painfully-slow development pace and frequent, serious schedule overruns. If customers can't rely on Kray's scheduling, then your talk of "future development" loses much of its value, because as customers we cannot depend on it arriving in any relevant/useful timeframe.

What are G. and you doing to improve Kray scheduling accuracy, and ensure it remains so?

Janusz Biela
09-11-2016, 12:30 PM
K3 will have his own physical materials - this is must (SSS, fast translucency, real displacement, new buffers, realistic glass, Ani Reflection, Oren-Nayar, and many others). Without working properly Core Engine G. can not start adding this. I wait for this moment because today without shaders is impossible render properly and we can not base on LW Node Shaders.
There, inside of K3 Core exist many features but still not activated because in this moment that is impossible.

K3 contains still old AA engine (also old Final Gathering) from K2 which is no so good. Grzegorz told me, the new AA will be available in K3.1 after finished Core. I dream to have AA Engine similar like in V-ray - there is really master piece of code!

Open beta K3 is only for me, rest users can test, complain, play, follow progress or forget about it. Bug tracker is closed with one reason: is just faster for me and G. in this moment develop K3.

K3 is developed since 5 years and just right now starting be visible part. Meanwhile G. had 2 years delay (life).
I think he managed quite good: good Photon Mapping engine you write around 2-4 years (if you are talented).

here is K3 SDK (updated continuously):
http://www.kraytracing.com/pub/docs/kray3_commands.html

jasonwestmas
09-11-2016, 12:51 PM
When you have two GPU top cards is not so bad. It is still slow as hell because there is no interpolation. It is really ok.

In what context are you talking about when you say there is no interpolation? Something like Redshift uses interpolation in its calculations because it is one of the first biased GPU render engines if not the only one that I've seen. I guess you might be talking about rendering with two cards together which I wouldn't do necessarily because the top grade cards work better by themselves. I like using one GPU for display and the other one is headless.

Janusz Biela
09-12-2016, 01:28 AM
I heard about biased render done by GPU. It must be very difficult to create code for that because GPU cards has a lot limits. Path Tracing Engines for GPU cards is easy to write because is simple.
GPU rendering in K3 will work in hybrid mode - it means some of calculations will do GPU some CPU.
I am not sure about speed render with biased GPU Engine, can be that it will be very slow, much slower than CPU.

When we talk about GPU, I am interesting in only with one thing: Interior render image with time render and info about GPU.
Is not working with me fancy menu, hundreds physical surfaces, million additional.

jasonwestmas
09-12-2016, 07:14 AM
I heard about biased render done by GPU. It must be very difficult to create code for that because GPU cards has a lot limits. Path Tracing Engines for GPU cards is easy to write because is simple.
GPU rendering in K3 will work in hybrid mode - it means some of calculations will do GPU some CPU.
I am not sure about speed render with biased GPU Engine, can be that it will be very slow, much slower than CPU.

When we talk about GPU, I am interesting in only with one thing: Interior render image with time render and info about GPU.
Is not working with me fancy menu, hundreds physical surfaces, million additional.

Thanks very much for your replies. Very interesting technology you are working with. Anything that allows the GPU to work with other onboard hardware is super interesting to me since GPUs are indeed limited in some respects. I just talk about redshift because it is the first GPU render engine that interested me because of its ability to move geometry and texture data onto the onboard ram when it exceeds the video card's capacity. It's slower that way of course but supposedly works well with faster CPUs.

Janusz Biela
09-27-2016, 02:17 PM
I just refresh topic because we go to K3 Open Beta 4 soon. Fresh update with high optimization in Core + update Anti Aliasing engine (still old from K2 but in K3.1 will be new)

K3OB3 1:49min:
134607

K3OB4 (number is not change yet):
134606

LW Native:
134605

...sorry I do not know well LW Native. I tried for both same setting with 10 bounce light in QMC system.

K3 VPR in action:

https://youtu.be/Dwn3M91_tDY

MichaelT
09-27-2016, 03:04 PM
I just refresh topic because we go to K3 Open Beta 4 soon. Fresh update with high optimization in Core + update Anti Aliasing engine (still old from K2 but in K3.1 will be new)

K3OB3 1:49min:

K3OB4 (number is not change yet):

LW Native:

...sorry I do not know well LW Native. I tried for both same setting with 10 bounce light in QMC system.

K3 VPR in action:


How have you set the scene up??

It only takes me about 2 minutes to render something similar in native LW render.

134608

rustythe1
09-27-2016, 04:50 PM
with no aa it would take me a few seconds, even with blurry reflections, vpr seems very slow too, that's the sort of speed I would get from a full archviz scene that had 10 million polys or so

jwiede
09-27-2016, 08:53 PM
How have you set the scene up??

It only takes me about 2 minutes to render something similar in native LW render.

134608

In fairness, the KrayOB3/4 example clearly has blurry reflections enabled on the floor and pillars. The pillars' surfacing also looks more complex (possibly a material) in the Kray examples.

There isn't much point in comparing times when the surfaces aren't visually equivalent. Janusz, can you post/show the scene (with LW materials) so we can really compare?

jwiede
09-27-2016, 08:56 PM
with no aa it would take me a few seconds, even with blurry reflections, vpr seems very slow too, that's the sort of speed I would get from a full archviz scene that had 10 million polys or so

I'm skeptical that you can get that quality of blurry reflections on that many surfaces from LW's renderer cleanly resolved in a few seconds.

Snosrap
09-27-2016, 09:29 PM
Thanks for the update Janusz. What I don't like so far from what you have shown is the unresponsive viewport performance when moving the camera around and the large steps of iterations the VPR takes. I'm thinking of working through material and surface setups - it looks like that will be really painful. You obviously have the materials already setup when you show these videos - I'm interested in seeing the performance while you're setting those up or experimenting with a surface - most of us don't have your material knowledge and will do a lot of trial and error until we get something that we like. :)

MichaelT
09-27-2016, 11:48 PM
In fairness, the KrayOB3/4 example clearly has blurry reflections enabled on the floor and pillars. The pillars' surfacing also looks more complex (possibly a material) in the Kray examples.

There isn't much point in comparing times when the surfaces aren't visually equivalent. Janusz, can you post/show the scene (with LW materials) so we can really compare?

Both his and mine are of similar complexity. But mine also have 4096 levels of AA, DOF, and soft angled lighting as well as 16 levels of bouncing. And all objects have the same material.

Here is another:

134612

djwaterman
09-27-2016, 11:50 PM
Very nice, the smooth blurry reflections is a really important thing.

MichaelT
09-28-2016, 12:04 AM
Very nice, the smooth blurry reflections is a really important thing.

Thanks :)

@jwiede: I agree. The best would be if he posted a scene we could look at. Because I am very curious as to why such a simple scene is so slow for him in LW (native).

rustythe1
09-28-2016, 01:21 AM
134613

yes, a quick setup at the same res, and 8.9 sec for a standard reflective surface with blur, and then just over 30 sec for conductor material with blur turned on, vpr almost instant,

samurai_x
09-28-2016, 01:31 AM
134613

yes, a quick setup at the same res, and 8.9 sec for a standard reflective surface with blur, and then just over 30 sec for conductor material with blur turned on, vpr almost instant,

That's much better effort. The kray render looks like it has sunsky system which makes it look vibrant and more dynamic. Contact shadows are also better.
Anyway the scene is too simple to make significant comparisons.

Janusz Biela
09-28-2016, 01:43 AM
Please download scene from here:
http://www.kraytracing.com/phpBB3/download/file.php?id=7362

I do not know native LW so maybe I setup something wrong. Also important is to keep 10 bounce lights in QMC system. I do not make tests in raytracing.

Janusz Biela
09-28-2016, 02:03 AM
Thanks for the update Janusz. What I don't like so far from what you have shown is the unresponsive viewport performance when moving the camera around and the large steps of iterations the VPR takes. I'm thinking of working through material and surface setups - it looks like that will be really painful. You obviously have the materials already setup when you show these videos - I'm interested in seeing the performance while you're setting those up or experimenting with a surface - most of us don't have your material knowledge and will do a lot of trial and error until we get something that we like. :)

The VPR is not developed yet at all. This what you see is just view port without any features. We can not make it working properly because simple K3 Core is not finished yet. So K3VPR is just extra future information which I can show.
In next updates will be with progressive scanning in K3VPR (also G. is tired from lags - VPR has only 4 progressive steps now...but as you know is so many more important things to do) Also AA phase is missing.
But what is important is that the K3VPR has 100% feature support of standard K3 render.
Actually I use only K3VPR - the pipeline system is staggering! You do not need even fancy previewer for materials (like other programs has) because EVERYTHING is in Real Time or will be in real time. You see everything is your designed light/GI conditions.

Janusz Biela
09-28-2016, 02:12 AM
Here is the extreme version without nodes. Simple 100% reflection and high blur.
134615
http://www.kraytracing.com/phpBB3/download/file.php?id=7363

djwaterman
09-28-2016, 02:18 AM
Thanks :)

@jwiede: I agree. The best would be if he posted a scene we could look at. Because I am very curious as to why such a simple scene is so slow for him in LW (native).

Actually I was referring to the Kray video, but yours looks nice to.

MichaelT
09-28-2016, 02:55 AM
I see now why it was so slow.. You disabled interpolation in the rendering. As to why you did that, I can only guess. You not only make the rendering significantly slower, it looks worse too.

S0nny
09-28-2016, 02:57 AM
Thanks Janusz for the update.
Here's a very fast test, I'll try with my other xeon render machine, it's about 3x faster than this i7 one, so consider ~60 seconds.

134616

This is probably not the best scene to do some comparison, but a couple of observations:
(I'm talking about LW native here)

- this is interpolated MC, in my opinion is interesting to get the better result no matter how we do it. Using non-interpolated MC gives just better and tight shadows and a fix noise/splotches for animation, there's no other considerable advantages in shaders look etc, expecially in simple a scene like this. It'd be better a simple interior to compare QMC/Non Interpolated LW native MC. I guess the differences will be much visible there.

- anything over 4 bounces in lw native is basically waste of render time, it doesn't provide any visible result. And even though, everything can be done in a much better way in tone mapping in post.

- Ray Recursion Limit = 10 in this particular scenario doesn't matter, because there's only one shader in the scene and it has its own limiter set to 3 using the logic node and bounces. using 10, 100 or 1000 it always be 3.

edit:

Add just for comparison, K2 default settings display srgb:

134617

Janusz Biela
09-28-2016, 03:32 AM
Yes.
Of course This is not scene for comparing render engines - is just too simple and contain only reflection with maximum 2-5 GI bounce probability light maximum.
Only can be test reflection/blur system in real full GI conditions. Let say true: today minimum system render is Photon Mapping + unbiased sampling (for example Corona render). NOBODY created enough good Final Gathering sampling which looks similar to unbiased sampling (QMC. Path Tracing, MTL).
So from me only full "throttle" tests: 10 bounce lights in PM system + unbiased sampling (QMC)...or we will see what G. will invent with new Final Gathering in K3.1...

MichaelT
09-28-2016, 03:48 AM
But if you want to look at final gathering, you should at least use that in LW too, and not Monte Carlo.
FG is faster, but less accurate however.

MichaelT
09-28-2016, 03:58 AM
Btw.. here is a good resource for looking at radiosity. I'm sure you know this well enough :) But I just wanted to bring this out if anyone reading this finds it useful. It's for an older version of LW, but works fine in later versions.

http://www.except.nl/en/#.en.articles.79-except-lightwave-96-radiosity-guide

Janusz Biela
09-28-2016, 06:15 AM
But if you want to look at final gathering, you should at least use that in LW too, and not Monte Carlo.
FG is faster, but less accurate however.

I use same system in both: QMC. I wrote before this is today minimum. In Final Gathering I would have much faster render.
I do not use interpolation for GI sampling in renders because this has no sense today.

Janusz Biela
09-28-2016, 06:22 AM
...I need also help from you all with LW settings for QMC system (good settings).
Here is next killer scene - but soon I will post here normal interior.
134618
134620
134621

Inside .lib files for changing surfaces.

Mastoy
09-28-2016, 07:32 AM
Tried to roughly match your renders in Octane :

134622
Rendertime : 5s


134623
Rendertime : 49s

Pathtracing, I let the render go until there was no noise anymore

MichaelT
09-28-2016, 07:43 AM
I use same system in both: QMC. I wrote before this is today minimum. In Final Gathering I would have much faster render.
I do not use interpolation for GI sampling in renders because this has no sense today.

"no sense"?? Why? I am genuinely curious as to why you think that.

Janusz Biela
09-28-2016, 08:46 AM
"no sense"?? Why? I am genuinely curious as to why you think that.

Do you see any render engine with full Final Gathering which is very accurate?...maybe V-ray but also I am not sure about the speed there - most of people use simple QMC.
It is always THRESHOLD when using Final Gathering starts be unprofitable to use. When render in FG takes 40 min and in QMC/PT 60 min - I choose simple second options because of GI Quality without errors.
In this moment do not exist good FG engine which really looks good. I do not talk about LW Native (which by the way has really good FG - way better then K2) but we know what is the quality of that...very artificial because of poor transporting photons - for me (no offense) LW interior renders looks like death - no life no feeling of real GI solution no realism.

Lately I just give up Final Gathering - no future until moment when will be new solution for that.
Even in popular Corona you have minimum: unbiased sampling in second phase. Is very long time render (really slow) but GI is staggering. This is way where to goes K3.
Of course you need also surface shaders system as complete solution, but for this I will back later when we will have our system in K3.


Here it is what I talk about:
The HQ GI solution with unbiased system (in this case QMC) based on PhotoMap system (Photons from light sources - very accurate photons) with VPR system in real time.
This is medium quality, almost Path Tracing quality under 10 min:
134624

Today you have only two choices:
- pure Path Tracing - go to GPU rendering or Maxwell
- PM + "QMC" - go to Corona

Nobody "plays" with FG today (except animation system)

Janusz Biela
09-28-2016, 09:48 AM
Anyway I will back later because I see bug in K3 with huge noise in reflection blur. At least G. fixed this crazy slow Anti Aliasing (is really fast even when we use still old engine)
Thanks for comments!

MichaelT
09-28-2016, 01:30 PM
Do you see any render engine with full Final Gathering which is very accurate?...maybe V-ray but also I am not sure about the speed there - most of people use simple QMC.
It is always THRESHOLD when using Final Gathering starts be unprofitable to use. When render in FG takes 40 min and in QMC/PT 60 min - I choose simple second options because of GI Quality without errors.
In this moment do not exist good FG engine which really looks good. I do not talk about LW Native (which by the way has really good FG - way better then K2) but we know what is the quality of that...very artificial because of poor transporting photons - for me (no offense) LW interior renders looks like death - no life no feeling of real GI solution no realism.

Lately I just give up Final Gathering - no future until moment when will be new solution for that.
Even in popular Corona you have minimum: unbiased sampling in second phase. Is very long time render (really slow) but GI is staggering. This is way where to goes K3.
Of course you need also surface shaders system as complete solution, but for this I will back later when we will have our system in K3.


Here it is what I talk about:
The HQ GI solution with unbiased system (in this case QMC) based on PhotoMap system (Photons from light sources - very accurate photons) with VPR system in real time.
This is medium quality, almost Path Tracing quality under 10 min:
134624

Today you have only two choices:
- pure Path Tracing - go to GPU rendering or Maxwell
- PM + "QMC" - go to Corona

Nobody "plays" with FG today (except animation system)

Thanks for the feedback :) Please don't take this as me trying to convince you into anything. I was genuinely curious. It is always interesting to hear why someone have the position they have. If anything, those answers help me.



Anyway I will back later because I see bug in K3 with huge noise in reflection blur. At least G. fixed this crazy slow Anti Aliasing (is really fast even when we use still old engine)
Thanks for comments!


You're welcome :)

S0nny
09-29-2016, 12:46 AM
...I need also help from you all with LW settings for QMC system (good settings).
Here is next killer scene - but soon I will post here normal interior.
134618
134620
134621

Inside .lib files for changing surfaces.

I'll give it a try: how do I use this .lib files to change the surfaces?

Janusz Biela
09-29-2016, 01:59 AM
.lib file you use in this way: right mouse button on top level of surfaces (in this case: Cornell):
134635
and choose "load Library" or "save Library"

By this you can load/save all surfaces.

Apparently I can not make real tests in K3 yet because still Core contain many bugs...but it will come time :devil:
At least we start bug tracker 2 months ago with this:
134636
134637

and now I can handle normal scenes....

The real purpose of K3 is system render with:
- first phase: Photon Mapping in PhotonMap system (when photons are shooting from light sources - this gives Path Tracing accuracy for photons and needs really ultra fast modern engine...in this moment only K3 can handle this in whole 3D market)
Here is my record - 1.000.000.000 photons received under 90 sec:
134638
- second phase: QMC sampling (no interpolation)
This system works perfect with VPR because you need shoot only once time photons and later you change only camera view and wait for GI sampling. System is around 10-20 times faster than Path Tracing with almost the same quality.


This all exists 10 years ago in K2 but engine was simple to slow for that (shooting photons phase very slow). Actually Corona contain something that: approximately PT photon mapping (interpolated Path Tracing photons) + QMC or MTL (I am not sure). But what gives Corona real boost is surface shader system.

S0nny
09-29-2016, 03:02 AM
This is the first scene, the standard non reflection setup.
It's MC interpolated with very tight settings, which is almost the same as MC non interpolated in this test. I'd say "visually" like 2% difference.

134639

To match the illumination I had to push the poligon light intensity about 3x times (I used 2000% in this case).

I'll try the reflection one also.

djwaterman
09-29-2016, 03:37 AM
This is the first scene, the standard non reflection setup.
It's MC interpolated with very tight settings, which is almost the same as MC non interpolated in this test. I'd say "visually" like 2% difference.

134639

To match the illumination I had to push the poligon light intensity about 3x times (I used 2000% in this case).

I'll try the reflection one also.

Where can I get the object files for this scene?

S0nny
09-29-2016, 03:40 AM
Where can I get the object files for this scene?

Is the same object Janusz posted here: http://forums.newtek.com/showthread.php?150674-Kray-official-announcment-is-coming&p=1484428&viewfull=1#post1484428

S0nny
09-29-2016, 04:22 AM
Here's the reflection one.

134641

This one has a much different look because the way reflection blur is calculated, I guess.*
Light intensity is 600% as in Janusz K3 scene, no need to push it over in this case. Only changes are Directional Rays = OFF, because it destroy radiosity calculation times with this number of bounces (and gives a totally different look also). Added more AA to match similar noise as K3.

I think it'd be better to use some energy conserving materials in both cases to have more solid output.

*IMHO lw reflection blur is too strong before 50% and shows minimal difference over 50% to 100%. Lately I prefer to use the POW node with a 1.5 multiplier which gives me a better control over reflection-blur.

Janusz Biela
09-29-2016, 04:28 AM
Please if it is possible, make tests with Native Lightwave in full GI solution: at least 6-8 bounce lights and no interpolated sampling. I do not know LW native (last time I used 10 years ago) and it can be something wrong with my settings.
Apparently last K3 update has bugs with GI noise and I have to make step back to K3OB3 (I posted all new bugs on bug tracker)
The reason of using QMC is that this is now the minimum standard in rendering 3D.
Using Final Gathering in simple scenes has no sense because in complex scenes amount of errors drastically raise up and in some situations it can be simple not profitable to continue with this system.
Apparently LW Native has very god FG (better then K2) but this is not solution for high quality renders (except animation when FG is the MUST solution).
I will post free and simple ugly interior here which I use for testing but more important is that it`s contain everything what should be tested and gives good average render time for small project:
134643
134644
134645
134646
134647

S0nny
09-29-2016, 06:07 AM
Please if it is possible, make tests with Native Lightwave in full GI solution: at least 6-8 bounce lights and no interpolated sampling. I do not know LW native (last time I used 10 years ago) and it can be something wrong with my settings.


The tests I posted are Lw native, Monte Carlo interpolated with 10 bounces (it's not Final Gather), you can try it on your other scenes, it should be fine.

What I did is just maximize some of your settings to get a MC non interpolated quality with a MC interpolated render times, so keep in mind this are very tight settings, they are reasonable but absolutely not optimized for speed or a speed/quality ratio.
MC interpolated in this case shows negligible difference from a non interpolated MC, but non interpolated can be like 50x and more slower.*

It all depends on what you are looking for:

-if you want to see the Lw native MC non interpolated quality, this settings gives you that quality, but with interpolated render times. Which make sense in my opinion.

-if you are looking to compare just MC non interpolated render times, well, there's no much that can be done in LW native, just add RPE, or AA or Shading/Lights Samples, or a smart balance between this parameters, as much as the noise go away. You'll get - in this particular scenario - nothing, quality speaking, but you'll gain a magnitude of render time.

*50x slower is not an exaggeration. Consider this: in Lw the MC non interpolated render is a choice for exterior render with very low bounces (2) or in some animation to avoid splotches, radiosity baking etc, and for distributed rendering to get a consistent uniformity between frames. In other scenarios like interiors and similar complexity, the time to clean the noise it's just unreasonable most of the times.

Note aside: some settings are resolution dependents, like Minimum and Maximum pixel spacing.

Here's the MC non interpolated scene.

Janusz Biela
09-29-2016, 10:47 AM
Here is the real purpose of PM+QMC in K3 (maybe I re-post it):

Pure Path Tracing (10 bounce lights - with less bounces will be faster of course) 3min 30sec:
134655


PM+QMC, 9 sec:
134654

Renders from K3VPR