One more note: "Low-priority" setting in Hardware Settings is real-time. You don't need to click "Apply HW settings" when changing priorities. Just change the checkboxes and it will arrange priorities on the fly. Will do the same thing for applying render GPUs too, but not soon.

Btw, the stats of CentiLeo iterations are adjusted a little bit. Before 0.51 the first iteration was progressive noisy thing. It wasn't included in render image (except for brief 2 iteration renders). But it was accounted in stats which led to some confusion during comparisons.
From now on each single iteration of CentiLeo render settings corresponds to 64 samples per pixel. Same is true for the first iteration. Progressive iteration render things are not accounted anymore in the stats and final image which has at least one iteration completed. This progressive iteration is just for display purposes to see the scene while user moves there. It takes 5-15 sec usually and has not multi-GPU btw The rest render process is perfectly multi-GPU.
In this design the increment of CentiLeo sampling is 64 samples. No way to set e.g. 16 or 32 samples per pixel. So keep it in mind during comparisons. CentiLeo has advantages for more complex renders which require some reasonable sampling, say 10 iterations or more.

Hello everyone!

CentiLeo 0.51 for Cinema 4D R16-R18 is available! At this time we have made the integration of CentiLeo with Cinema 4D tighter! Support almost everyting from Texture Tag mapping including tiling, offset, repetitions and alpha-channels of CentiLeo materials. All this can be used to combine the stack of CentiLeo materials with semi-transparency of materials and geometry (if we are on the ground-level material). Just like in Cinema 4D. Mappings like Cubic and others are in TODO list. Also have added Variation and Random shaders. See the full list of changes below.

Download: http://centileo.com/forum/forum18/126-download-link-centileo-for-cinema-4d-v0513-alpha

Change Log: cntlc4d 0.513 alpha (2017 August 23)

[C4D plugin] Fixed a bug in viewport preview texture render that could cause crashes of Render Queue.
[C4D plugin] Fixed bump strength range (now it’s arbitrary positive and negative).

Change Log: cntlc4d 0.512 alpha (2017 August 19)

[C4D plugin] Fixed double scene evaluation bug which could cause slight IPR slowdowns .
[C4D plugin] Fixed: new internal representation of materials and shaders for 0.51 is created automatically (no need to start render for this). Without this before user starts a render there could be confusions in material/shader copy/paste operations.

Change Log: cntlc4d 0.511 alpha (2017 August 13)

[C4D plugin] Fixed issue with Subdivision Surface. IPR renderer gets “Subdivision Editor”
setting, Picture Viewer renderer gets “Subdivision Render” (final) setting (not from Editor).

Change Log: cntlc4d 0.510 alpha (2017 August 12)

[core] Faster GPU initialization stage for Multi-GPU.
[core] Added low-priority setting for GPUs (see Hardware Settings). They are ON by default in IPR mode for ALL GPUs. For fluent work in IPR mode it’s HIGHLY RECOMMENDED to keep enabled “low-priority” setting for DISPLAY GPU and switch if OFF for other render GPUs which are not connected to the monitor. Low-priority option is useful only for display GPU because it is used for Viewport display. If you have many GPUs then display GPU can be excluded from IPR render.
[core] Added Variation shader. It generates different colors using HSL color model if material with this shader is assigned to different objects. Each HSL component has the range which is divided with a setting “resolution” and in this range a random color is picked automatically for different objects. Objects are considered different if they have different names. Later will add special object tags for this purpose to extend flexibility.
[core] Added Random shader. It assigns random shaders for different objects (max 10 shaders to select one random from it) if some material with this shader is assigned for different objects. It’s possible to construct many random shaders on top of each other. Later will add Random Bitmap shader.
[core] Removed Geo Mask slot from Multi Material due to more flexible and quite fast new Alpha channel. Maybe will return it later
[core] Added Alpha channel (same as Opacity) for Standard CentiLeo material. It’s stochastic (but very fast) and can use combinations of shaders. However, the depth of effect is limited to “max ray bounces” setting.
[C4D plugin] Added support for material stacking using standard Texture Tags. This feature understands selection tags, material Alpha channel, offset, tilling, repetitions and up to 4 UVW Mappings from the row of object tags. The UVW Mapping that is assigned to material shaders is the next right one to the Texture Tag. Default mapping of material (using this Texture Tag) and tilling parameters are propagated to internal material shaders (like Bitmap and etc.). This works as suggested by default Cinema 4D behavior with a limitation to 4 UVW Mappings per object. Other procedural mappings like Cubic, Cylindrical, etc. are in TODO list.
[C4D plugin] Each of the CentiLeo shaders such as Bitmap, Noise etc have “default” mapping parameter which inherits the mapping of the entire material which is enabled in the Texture Tag. Each shader like Bitmap can change this mapping and relative tilling too.
[C4D plugin] Fixed: when materials linked to Selection Tags are inherited from some parent object to child objects then polygon Selection Tags of child objects are processed correctly.
[C4D plugin] Fixed: when CentiLeo material is assigned to an object then default projection is UVW.
[C4D plugin] Fixed: when CentiLeo material is copy-pasted then all the shaders are also copy-pasted with it (not as instances but as copies). It also fixes the problem of copy-paste to another scene.
[IPR] Fixed non-working IPR in C4D R16.050 build.
[IPR] Polished few things that caused the random stalls.
[IPR] If IPR window is minimized and not present in the task bar then click “CentiLeo IPR” command in top menu again to get it back immediately.

Known issues to be solved

- Displacement mapping feature is deactivated due to large code changes overall. This feature will be recovered within the next few short CentiLeo updates.
- May not initialize if AMD GPU is in your computer together with NVIDIA GPU.
- Very limited scene converter tool from Cinema 4D native materials (only diffuse color and texture are converted and translated if assigned to the object)
- May not support non-latin texture paths.

Important note

We are currently using JUCE library (https://juce.com/) for IPR GUI under Personal license (which is free) for initial testing purpose. In IPR window it will show in the right bottom corner a splash screen for the first start and it dissapears in few sec. It may also connect to Google Analytics to collect end-user stats. Juce guys claim this is nothing bad but stats collection so they know who is using them. However, the plugin works perfectly without any Internet connection.
If everything works well on main technical side then we will buy the full license of JUCE and get rid of these small issues.

Please test it and give us feedback to discover further improvements!

Best regards,
Kirgman

Hi Daniel, yes of course.
0.51 will come in around 2 days but with no 2018 Max. Hope to do it in 0.52 release

Hi Rodrigo! I have made further entire tests with Sculpt Tag. It is just silent if we check it's results in a usual way. It doesn't report that geometry is changed. Currently only Manual Reload option for this in IPR. That's dissapointing and it should be a bug of Cinema or some plugin that they have acquired. We posted to them about this. Hopefully they will suggest a workaround. Other renderers with IPR also have the same problem. Btw nobody before us asked Maxon about this on their development forum.

As for Material Editor scene then it is hardcoded in source codes just for better speed. I.e. no actual scene in file. There are plans to extend this with preview models other than just a sphere. E.g. there can be boxes, planes, sphere, cylinder, torus. And probably there will be a centileo shaderball there too (not sure yet).
The light source in that hardcoded scene is disc area light (quite small) that was tuned to get less noise for the case of very strethed anisotropic highlights. When we add GGX (soon) then things may change. Anyway these ME works are postponed untill we improve our core with features for more mature state (versions 0.6 and 2-3 months since now).

I can type params and coordinates of that scene just to recreate it in C4D file after we finish 0.51.

I have evaluated the SculptTag issue with IPR. I have found that if SculptTag is directly attached to an object then updates don't perform in IPR. But if not attached and we modify some object with e.g. a knife then changes are applied automatically.

However, the issue seems to be a potential bug of SculptTag. CentiLeo IPR relies on some special dirty flags that signal that something in scene has changed, we go down and look at object dirty flags, then tags, and etc. And only if dirty flags are on we get the data content to load to IPR. And when SculptTag is attched no changes are reported. Nowhere. It shouldn't be so. I think it should be the same issue with other renderers, but I will check with Maxon techsup of course.

Quote
Rodrigo Bitencourt Rodrigues wrote: Bug: Centileo should avaluate the play and stop(please add a stop button) of the IPR, not the window. because sometimes I need render in Picture Viewer but I want to maintain the IPR opened(not rendering, just opened).

If we open Picture Viewer then we need to clear the scene of IPR from memory (can't keep 2 scenes in CentiLeo since it is a lot of vaste of resources).
If we implement maintaining the IPR window and open PV for work then IPR will be just a window with cached image and no scene, no interactive render process while PV is running own render.

Hi Rodrigo! Thanks for bugrep!
It seems I can reproduce. Sculption layout? Not tag (haven't found it). It has started to update IPR as expected, then at some point stopped reacting for updates.

Quote

Thomas DION wrote: Hello, great work on debugging and optimizations ! I was wondering if it would be possible to rename the render pass "Reflection 1" and "Reflection 2", as they now are exported as "RenderName_.Beauty Reflection 10060.jpg" and "RenderName_.Beauty Reflection 20060.jpg" for example, and so are read as frame numer 10060 instead of 0060 in softwares like After Effect and such. Thanks ! Best regards, Thomas

Hi Thomas, of course you are right! Will fix that. Next version should be out in few days.

Quote
Rodrigo Bitencourt Rodrigues wrote: Question: Are in the roadmap a separated config for preview and final rendering? I think its very important IMHO. I mean, not only rendering configs but hardware configs too.

Never thought about it. Need to think how to do it without confusion for user. Think that the only potentially good candidate for this is Hardware settings and maybe sampling params.

Hi Tiago, thanks for the test! It seems that in this setup we are 25% slower than their BF-BF (our's is only BF-BF) and probably very slightly less noisy. And for this scene 3.5x slower than irradiance and photon methods. I prefer not to implement these methods due to their limited application and potential strugle with parameters/memory for tiny objects which will be very important for us in the future. Instead we prefer to concentrate on our BF optimizations.

However, in this test there are some aspects needed to be mentioned:
1) Would like to recommend to increase the light of the background yellow room for Centileo case. Otherwise the proportion of lighted and shadowed areas is a bit different.
2) Double check the number of GI bounces in RS and number of ray/diffuse bounces in CentiLeo. For 1:1 comparsion they should be all equal.
3) In CentiLeo the first iteration (from 0 to 1) is used for interactive real-time navigation only (single GPU only). It's fast but only for preview. It's noiser than the higher iterations and produce an overhead for several (maybe 10-15 sec). If the number of Min/Max iterations is 2 (like in this test) then the first iteration is partially visible in the final frame (and it has more noise than can be). If you put Min iters = 3 (at least) and Max iters higher than 3 then the first iteration will be excluded. I will probably fix this and put the minimum value for Min = 3 )
4) One iteration of CentiLeo corresponds to 64 samples.
So to make more close comparisons I would recommend to align Redshift samples to 64 increments. E.g. if you make Redshift min = 128 and max = 512 it corresponds to CentiLeo's min = 3 and Max = 9 (recall that the first iteration is not used at final renders). And play with noise level value.
5) CentiLeo's adaptive noise driven sampler is more apparent if we run reasonable amount of iterations (like 5, 6, 7 or better tens). At low number of iterations it is almost useless.

6) Why we did it in such a way: simply because we wanted a fusion of progressive - bucket mode with gradual fusion from 0 to 3 iteration in IPR mode which we bet on. And we want to make quick buckets and iterate, iterate, iterate many times like in progressive mode where user can see the whole picture at any time. We also can't make internal bucket size less than 64 samples per pixel because 64 is a very good balance for final render speed coupled with some features. We also preferred to optimize on some more complex cases which naturally require more sampling and etc. etc. When compare on small number of samples/iterations then CentiLeo shows all the overheads that come from this design and can be slower compared to others for simple cases (but still pretty fast). However, these overheads are negligible for more complex scenes and has impact on simple cases.

Why in CentiLeo and Redshift the camera pos is different? Have you moved the camera or applied some offset or etc.? Or our bug?

These things will come in the nearest 1-3 releases. As for IPR then indeed it takes more CPU cycles than before, refresh rate is too high (need to reduce it).

Quote
S M wrote: Thanks! What about alpha visibility? Objects with refraction are solid in alpha. Am I missing something?

This is not yet supported. Must do it. Partial alpha-channel output is currently working for shadow catcher shadows.

My bro have shown me some complains of Tiago about CentiLeo speed in interiors on your FB page. Show this to Tiago:

Pre info: CentiLeo's: Min/Max iters 3 / 17 is like 128 / 1024 samples per pixel. In both engines noise limit is 0.01. However, CentiLeo's noise limit of 0.01 seems to be like 0.003 - 0.005 of Redshift. In both renders 4 GI bounces were used and diffuse material with 80% bright white. Light tuned to be similar.
In one case there are area lights in windows. In another case just white solid environment without portals.
Just so simple classical interior. 2 GPUs 970GTX (secondary, not on display).
In same time CentiLeo produces less noise on same HW on same scene. Sometimes not so much less. I have other scenes where in shadow areas CentiLeo has more advantage. With more geo also more advantage.

But I think we may still have some problems on too small GPUs (complains on speed / IPR lags only from them). Will look at this, but not priority. More features is must.

Quote
Rodrigo Bitencourt Rodrigues wrote: + Area Light scale should respect the size of the Area light DETAILS TAB(Size X and Size Y) that way I can edit the size via viewport with Area Light helpers..

But it respects! So do for inner/outer angles of spot light. There is additional are scale in light tag too which multiplies area on top of details tab.

Quote
Rodrigo Bitencourt Rodrigues wrote: + DOF Sensor Width should respect Camera F-Stop value, like a real camera. lower F-stop = more DOF. high F_Stop = less DOF

ok, should do a lot of things for camera

Ok, I see. I will try to double-check and polish some things to make better stability for new IPR implementation. Strange because in Max plugin it doesn't do anything serious when close the window.
Also will add GPU priorities to Hardware settings in attempt to improve IPR fluence when there is just 1 GPU in a system.

The manual is old enough. Need to upgrade it a lot. Will do it after some 0.55 version when will add more shaders there.

Quote
S M wrote: sampling params are different now as in manual. What is the best approach to speed up rendering and to fight fireflies?

Min/Max iterations is like number of samples. Each complete iteration makes 64 samples per pixel for entire image. If the status bar shows partially complete iteration then it means the current pass over image is partially done. The first iteration (from 0 to 1) is progressive for navigation purposes only during interactive work with the scene. The other iterations are much more clever and are tuned for less noise and final quality, hi-res.
When the sampler reaches Min number of iterations and desired "Noise level" for pixel then it stops rendering it and contiunes processing only pixels which have more noise then specified in "Noise level".
But if pixel sampling has reached Max iterations then it stops regardless of noise value for pixel.

Redshift, Corona, Fstorm all have similar settings. Octane too (but they have slightly more settings).
Our noise level internal estimation is conservative. At value 0.01 it doesn't give chance for noise where e.g. Redshift stops at 0.01. This is good for quite dark shadow areas.

The other optimization settings are as usual: number of all ray bounces (and max number of diffuse only bounces which can be lower than overall num of bounces). Diffuse bounce is a ray bounce off surface with roughness higher than 0.1.

We also have clamp values
1) for direct lighting,
2) specular direct lighting (everything reflected from perfect mirror or glass with roughness less than 0.1),
3) and clamp for indirect light.
The last 2 may give fireflies in some cases. Try reduce them and see.

What kind of scene you have fireflies?

Quote
Sergej Mostnar wrote: Very nice. Sometimes Max stops responding after ipr or material window closing. Have to force-close app.

Hi Sergej, thanks for telling about this.

Pls, can you describe this case? What GPU (or several), what kind of scene and the actions you did before it stopped responding.
I had some crash on my computer with 2GB GPU (750ti) which reported just 1GB remaining for renderer after OS and viewport consumed their GPU resources. This remaining 1GB is quite small for renderer (it can work but not as efficient as with 4GB+). How much memory available for GPU can be seen in Harware Settings of CentiLeo for each listed GPU.

Do you close the whole 3ds Max or just IPR to make it alive again?

This is the first version with new IPR engine and need to find all these bugs.

Hi! Well, unclamping Weights from 1 is non-phys, but who stops us from this. Right now it is possible to override the weight values (or even Color) with const shader with arbitrary value.

As for IPR then together with the list of active cameras the plan is:
183) IPR: progress bar (scene compile, IPR scene reader, texture parser, render, region), PCI Express BW per GPU, pixel inspector, tabs DoF and Light pass, white balance picker, image history, A|B comparison, render stamp, more post stx. Almost a picture viewer

Reaploaded with small glass fix: shadows casted by glass objects look more proper

Hi everyone,

CentiLeo 0.50 for 3ds Max 2014-2017 is ready! It has a lot of important bugfixes for refractions, diffuse interreflections, etc. Some things like new IPR GUI took a really a lot of time due to fixing new bugs related to it. But this was solved. Hope to add more graphics features to CentiLeo asap.

Download here: http://centileo.com/forum/forum13/124-download-link-centileo-for-3ds-max-v0500-alpha

Change Log: cntlmax 0.500 alpha (2017 July 19)

[IPR] Changed GUI engine for IPR mode to JUCE library (www.juce.com). All IPR controls like image resolution and post settings go to the IPR window. It opens the IPR window faster, process images faster and is very cool. Later will add DOF, light pass settings and more cool things there too.
[core] Fixed wrong glass refraction patterns. It is also accelerated a bit for many multiple internal glass reflections.
[core] Improved the look of shadow from the glass objects.
[core] Improved rough glass render speed and look.
[core] Fixed loss of light energy for diffuse interreflections.
[core] Faster render for HDR lighting with very bright high-contrast spots (thanks to a bugfix).
[core] Fixed problem when display GPU participates in multi-GPU rendering. Before fixing this the display GPU could even deaccelerate collective rendering. Right now display GPU accelerates the multi-GPU render (however remember that display GPU also shares resources with native C4D viewport).
[core] Reduced peak RAM memory consumption when scene with huge meshes is loaded. Just valid and useful data takes place but temporary data doesn’t consume huge memory like before. E.g. for 50M poly meshes it is reduced by ~10GB.
[core] Out-of-core texture cache doesn’t consume all maximum capacity upfront. Now it saturates the cache on demand (when you add textures until the limit that you set in max OOC texture cache setting). It results in minimum RAM consumption (2GB) for blank scene (lower than before).
[cntlmax] Fixed window crashing error when you have opened Corona render IPR before CentiLeo.
[cntlmax] Fixed crash when change animation time and cntlLight (or cntlUber) was present in the scene.
[cntlmax] Fixed zoom scale of viewport light source representation. Light source names are off by default (may be switched on in LS settings).
[cntlmax] Fixed RAM memory consumption when IPR is open for several hours.

Known issues to be solved
- Displacement mapping feature is deactivated due to large code changes overall. This feature will be recovered within the next few short CentiLeo updates.
- May not initialize if AMD GPU is in your computer together with NVIDIA GPU.
- Very limited scene converter tool from other renderers (only diffuse property and textures from Standard material and Vray material)
- UVW mapping change of parameters may not be accounted by IPR scene auto update mode. To be solved.

Important note
We are currently using JUCE library (https://juce.com/) for IPR GUI under Personal license (which is free) for initial testing purpose. In IPR window it will show in the right bottom corner a splash screen for the first start and it dissapears in few sec. It may also connect to Google Analytics to collect end-user stats. Juce guys claim this is nothing bad but stats collection so they know who is using them. However, the plugin works perfectly without any Internet connection.
If everything works well on main technical side then we will buy the full license of JUCE and get rid of these small issues.

Please test it and give us feedback to discover further improvements!

Best regards,
Kirgman

In 0.49 times I have promised to make TextureTag and Alpha channel for material stacking by 0.50. And we didn't make it by this moment due to so many issues that needed to be solved for robust IPR work under new GUI engine. It really took lot of time for polishing. However, we will put more efforts to make promised features very soon.