[thxraph]

@Aniki, well, yes, the "black dots" will disappear, it is just that even with that config, the render needs more time.

@numerobis: yes, this improvement isn't limited to SSS (if it is integrable caustics) it will make better & faster all the dielectric materials.

And hopefully, you'll see the same incredible side effects we seen with Arion when the BRDF get revised and the dielectrics enhanced: swimming pool caustics, yes. Literally, caustics seen throu dielectric without using any kind of algorithm like MLT.

Edit: About the material, yes, we made it match the Maxwell's playing with the REF0/90 grey values.

Raph

[tom]

Raph, first of all; thank you for contributing to this thread. The comparisons are very nice and valuable.

I'm pretty sure that it is what 2.6.1 intend to resolve, so i guess it's gonna be much better on the maxwell's side when it comes out. Then one will have to update the thread.

This is correct. Because, the current look is incomplete at this SL and it will take a lot of time. The noise comes from indirect illumination which is converging slow in this case. With the new update, this is not going to be such a pain anymore. Here's a small example showing an indirectly lit object:



As seen in above example, the current look of dragon becomes not only less noisy but also remarkably different by the improvement of indirect illumination. At that point, I'm trying to figure how to compare it to the current image produced by Arion. Because, as the current one matches Maxwell other than noise, does it mean Arion is missing something in quality in order to avoid noise maybe?

Therefore, I should post a image for leveling up this comparison but, I need to know the reached SL of your 60 min render.
Apart from that, I'd say maybe it's about different roughness models but, are you sure the scene illumination is same? The speculars look way off...

[Aniki]

Thanks for joining in, Tom;)

I'd love to see the dragon scene rendered with maxwell 2.6.1.

Can you also share any preview/info on the caustic through transparency ?

Cheers, really looking forward to the gifts, NL is holding ready for us this year

Aniki

[tom]

Maxwell Render 2.6.0.0 (60 min)


Maxwell Render 2.6.0.x pre-release (60 min)


Animated comparison between above versions (GIF anim):

[numerobis]

WOW! great improvement! thanks for posting Tom!

[Maximus3D]

Wow that looks a lot better in the shiny new Maxwell update Tom
Feel free to throw in more tests here

/ Magnus

[thxraph]

hello guys,

1H Maxwell dragon I rendered here (Core 7i 950) is SL 16.19 (glad I kept the mxi xD)

That is nice update, integrable dielectric caustics + enhanced rough dielectric, right?
The maxwell's scene is just the same as the arion scene, especially lighting-wise, this scene is lit by the little emiter, and this little emiter only (no env light, that would be cheating, right?).
I think that the Maxwell 2.6.1 dragon render match pretty much the way Arion is cleaning now.
There is nothing left quality-wise in Arion in order to clean the noise faster, just that this little more that you added to this incomming version of Maxwell is already here in Arion, that is the difference (+ GPU speed). Roughness-wize, I think matching the values from Maxwell to Arion (or otherwise) will produce a slighly different effect, yes. Not that one is more correct than the other, i think it is just a scale that differs a bit. We tested (in Arion) rendering balls from Roughness 0 to 100 and it is pretty much linear, but only a slight difference between the 2 engines is enough to produce a visible difference I guess, also, no 'force fresnel' in Arion, the reflectivity boost was achieved playing with colors are 0° & 90°.

What i'm sure because i've been able to see it by myselft if that the dragon we rendered in Arion has small geometry problems (missing faces), not much, around 5 i'd say, and that are the 2 darker dots you can see on the head (after checking).

Tom, if you have other scene that you can post so we can test, feel free to do it. There is one simple rule tho, the harder the case for an unidirectional renderer, the more Maxwell's & Arion's rendertime will match (without 2 or more highend GPU, ofcourse). The larger the emiter, and the rougher the material, the bigger the gap in render times (huge Arion speedup). I'm pretty confident that the dragon's scene demonstrate that Maxwell next update has caught up on Arion caustic-wize, yet the noise, much more uniform now is maybe still not as uniform as on Arion (maybe a different seed number will change that, i'm not technical enough at this point, this is just based on what my eyes see, and they are tired xD).

Raph

[Aniki]

Great improvement ! Thank you, Tom.

So in favor of the better cleaning of small areas where extreme noise is due, maxwell 2.6.0x will introduce a slight overall noise again ?

[thxraph]

@Aniki: no. No slight overall noise. Regular noise all over the SSS/diel material. It will clean better, faster and also (and that is the most important) will look better, as full internal caustics are integrated.

Raph

[David Solito]

@tom : Houaaw. Can I post the renderlinks on the french forum?

[thxraph]

Sorry Tom, stupid me, I didn't give away the last Arion's material I used in the render.
here it is:
http://gallery.realistic-design.com/ari ... ch_mxm.rcm

Raph

[tom]

That is nice update, integrable dielectric caustics + enhanced rough dielectric, right?

Actually, it was specific to SSS materials. If you remember frosted glass dragon, the ones you say are already working nicely for a long time.

The maxwell's scene is just the same as the arion scene, especially lighting-wise, this scene is lit by the little emiter, and this little emiter only (no env light, that would be cheating, right?).

Yes, additional lighting (especially diffuse) would be a cheat for sure. I just think your camera position is different, then. Because the images are not perfectly overlapping.

I think that the Maxwell 2.6.1 dragon render match pretty much the way Arion is cleaning now.

Mmm..now that's something I perceive differently when I pay attention. Would you like to check the below animated comparison? The yellow tinting in a positive direction and smokiness with the internals represent the material more correctly comparing to the greyish hollow tints of Arion and unimproved Maxwell version.

...also, no 'force fresnel' in Arion, the reflectivity boost was achieved playing with colors are 0° & 90°.

Force fresnel is something you can't always tolerate with LDR color inputs for Reflectance and especially setting an arbitrary balance among them. Because, Force Fresnel disregards the input luminance of colors and ensure the fresnel flow is forced to reflect the given Nd. That's not something cheap as it seems.

Tom, if you have other scene that you can post so we can test, feel free to do it.

I didn't have something on my mind for a comparison. Just saw the thread and jumped in. I just can agree this is not a sexy material and color combination to throw lots of tests around. When I have another, I will...

Sorry Tom, stupid me, I didn't give away the last Arion's material I used in the render.
here it is:
http://gallery.realistic-design.com/ari ... ch_mxm.rcm

I used the material that Aniki posted here. http://www.maxwellrender.com/forum/view ... 45#p351145

[tom]

Just to make sure, it converges to a noisefree state w/ enough CPU time... I wonder, after how much time would Arion catch the same look. Or, will it ever?

Maxwell Render 2.6.0.x pre-release SL 19.88 (3 hours)


Maxwell Render 2.6.0.x pre-release SL 23.00 (10 hours)

[thxraph]

Use the last material I posted.

About the color difference, something is odd. The very fact that the floor color doesn't match is weird already. I think it would be better to use a "normal" tonemapper like Reinhard or linear, so we can have the "same" results.

My scene has a slightly different camera angle, yes (that would need to be fixed, although it is not dramatically different).

About the force fresnel: ok, now i understand what it does. In Arion I would use plastic coating to produce the same effect/behavior. Nobody said it is cheap, but I wasn't sure of what it was for exactly.

No additional lighting, of course. The emiter is the only light source.

About the colors and smokyness, it's pretty much density coeficient related, their might be a mismatch between Arion & Maxwell (in fact, it is more than probable), let's see if you have better results with the last material, but i'm pretty sure it's a matter of settings, not of internal computation. Playing with the settings we could match Arion to Maxwell and Maxwell to Arion, but not sure it is worth wasting too much time of that. I like the one you achieved with Maxwell better tho. It looks denser, more colorful. Sexier to say it that way

The last animation shows that Maxwell, even if awesomly improved (i'd like to stress that!) still have brighter spots that cause some irregularity compaired to Arion. It is much better, yet not perfect. Don't know how it looks on a long run. Also, this Maxwell is still in internal dev/beta stage, while Arion is a release, so we can't really compare. Arion couldn't be taken as a reference here, the noise is just slightly more uniform to me, yet you know how it works, in 1years or maybe 2, devs will come up with a new awesome super-correct algorithm that makes all the olders obsolete, that will produce super uniform noise in no time, increasing CPU/GPU power helping.

About the material and not being very familiar to polyurethane, at first sight it makes me think of jellified chicken juice xD

Just to make sure, it converges to a noisefree state w/ enough CPU time... I wonder, after how much time would Arion catch the same look. Or, will it ever?

It will, but it is more a question of settings than rendertime to me. As the colors and how it will look clean is pretty clear after 25min on my computer. On Marco's (another Arion tester) computer, for exemple (6x GPU), 25min is the time he needs to have it outrageously noisefree.

Raph

[David Solito]

@tom : Houaaw. Can I post the renderlinks on the french forum?