Jump to content

Suppress small Artefact in PBR render


Recommended Posts

I have relunch a render with your advise dennis, it's in fact faster and i am now very close from Arnold

 

here are my settings

- full PBR with portal (no photon map)

- pixel sample : 8*8

- min sample : 2

- max sample : 8

- noise level : 0.008

- limite diffuse : 2

- limit reflection : 2

- min reflect ratio : 1

- env light sampling quality : 2

 

i think that when you have perfectly undertsand how noise can be control in Mantra you have a make a big step ... :)

 

compare_2.jpg

Edited by sebkaine
Link to comment
Share on other sites

Glad that its working out. 

 

You have to keep in mind though that the default mantra shader you are using has a glossy reflection component while Maxwell and Arnold seem to be diffuse only.

This is obviously putting mantra at a disadvantage.

 

EDIT: I just recalled that you wrote you are using the GGX shaders. These are indeed noisier than Houdinis builtin shading models.

Edited by dennis.albus
Link to comment
Share on other sites

Ok now with clarisse ifx 1.6 PLE.

it's a first test so i guess it could be optimised ...

 

i had a lot of difficulty to get rid of the noise in full path tracing , with occlusion light it's fast ,

but with only path tracing i had to go very high in the light sampling, i think i am doing thing wrong here but ... here are the Settings

720p => 40 min

 

 

- gi_monte_carlo

- affect diffuse only

- no occlusion lighting

- light samplig quality : 32

- max bounce : 2

- raytracer anti aliasing : 8

 

 

clarisse_006.jpg

 

i must confess that this first experience with clarisse wasn't that impressive

- you lost the nodal tree organisation to use a sort of very good outliner

- you lost the ROP tree organisation for a layering system

- the interactive preview is very efficient but houdini IPR while not as good is still relevant

 

well either from

- flexibility

- quality

- speed

- organisation of the workflow with geo / pass

 

i find that Houdini has a lot to offer , the big cons is that all this greatness is sort of hidden behind a pretty confuse

- mantra shader

- mantra rop

 

in this regards Arnold / Maxwell with their minimalist approach are absolutely great , you have what you need and just what you need

no more ... no less.

 

and i would say that maxwell BSDF layering system is my favorite one. it has the good part of the pixar slim delux + simplicity of arnold + stellar / unbreakable quality of the material. Switching mantra to a BSDF layering system instead of this monolithic mantra material could be an option ?

Edited by sebkaine
Link to comment
Share on other sites

Very interesting thread! The last weeks I started to dive into houdini in my freetime after some years of max and vray....vray is very fast with irr/lightcache combi but shouldn´t it be compared to the brute force/brute force method in Vray in this case? For stills it´s no problem but an animation with the vray settings above would cause a lot of flickering...at least for moving objects. Please correct me if Im wrong.

king regards 

Jon

  • Like 1
Link to comment
Share on other sites

I do want to also note here that in previous testing, on the same machine, Mantra on OsX is ~25% more inefficient than Linux.  That's one frame in every four!

Link to comment
Share on other sites

I couldn't resist to throw the stuff in RedShift

 

1080P => 18min

My card is a poor Ge force 660, so i'm quite sure that on a proper GPU like a 970 redshift beat V-Ray.

The quality is very good imo ... definitly a futur Rock Star ! and probably one of the fastest engine on the market ?

 

redshift_001_18m30s.jpg

Edited by sebkaine
Link to comment
Share on other sites

 For stills it´s no problem but an animation with the vray settings above would cause a lot of flickering...at least for moving objects. Please correct me if Im wrong.

Yes, irr map/lc is not the best combination for scenes with a lot of moving objects, but there are several workarounds which makes it more efficient than the brute force method. That is the reason why it is the best choice for arch-viz - very easy to setup and super fast.

 

 

1080P => 18min 

My card is a poor Ge force 660, so i'm quite sure that on a proper GPU like a 970 redshift beat V-Ray.

Well, there is a v-ray RT version that runs on CPU/OpenCL/CUDA , so check it if you have a chance. Have fun!

  • Like 1
Link to comment
Share on other sites

3Delight with it's killer path-tracer. Free license now goes to 8-cores, up from 6 :) It's meant to work in Houdini but we need some shaders written I'm thinking.  

 

HD 1920x1080 - 21min extrapolated out to 12 cores, (33min actual to the limit of 8 cores) on 2 x 3.33 Xeon 6 core.

 

* warning * 25 min experienced operator :)

 

Link to full res...

https://www.dropbox.com/s/uhxek3vr1kb2gkf/3Delight2.png?dl=0

post-8321-0-16179400-1425598532_thumb.pn

Edited by tar
Link to comment
Share on other sites

Cycles in Blender 2.73a OsX - tried the Gtx980 but failed. CPU; 17mins and 128mins - for some reason the outlines were added...

 

*warning 10 min experience with Cycles *

post-8321-0-14336100-1425612289_thumb.pn

post-8321-0-23347000-1425612387_thumb.pn

Link to comment
Share on other sites

Here it is my contribution:

 

1080 => 50 min render time on a 3GHZ quad core

pixel samples 3X3

min samples = 4

max samples = 32

noise level = 0.001

 

post-5743-0-13361300-1425614254_thumb.pn

 

 

PBR + Portal light (4 sample quality) with a (low res convolved HDR map) + Photon Map (filter samples 10/prefilter samples 500/distance threshold = 1) I am not sure if you really need to bump up the filter so high.

 

Similar to the previous images and less than 1hr render time, I think I would not have problems with flickering noise in a animation maybe a little bit on the aliasing, I've seen worse that would get away in a shot but I would have to see sequence render to be sure, decreasing pixel samples really helped to reduce render time a bit and I think that using a lower resolution convolved hdr map, which I did inside Houdini, also helped.

 

I still think there's room for optimization maybe get a little more alias definition without increasing to much render time, a 12 core machine would definitely help and also would be nice to use real shaders to see how much render time would really change.

 

I know it's a little bit off topic but I think that worth to mention that mantra has a built in network render where you can add machines to help render your frame locally, it's really great for look dev even more if you have a farm, you can have 10-20 machines rendering your frame at the same time in the renderview! your workflow for interactive look dev would definitely change, one more option that you can get from Mantra.  :)

 

I'll post more if I have more meaningful updates.

 

Cheers!

 

Update: My image was reduced automatically I don't know how to post the full image.

Edited by Mzigaib
  • Like 1
Link to comment
Share on other sites

@Jonathan => good point for the flicking that's why we now use maxwell instead of vray in my previous company. it's slow but if you have a big farm you know that you will have

- no flicking

- stellar quality

But i still consider V-Ray one of the best engine of the markett with the biggest scope in possibility for rendering algorithm.

I hope it will arrive sooner or later in H.

 

@Pavel => Thx i'm gonna check this

 

@Marty => I always love 3Delight when i was working on Maya it was my renderer of choice. it suffer for a long time of the same problem mantra has.

Great beautiful engine / Average awkward interface with the user ...

 

@Michel

Thanks for the tips i think that the arnold logic of 4 / 8 / 12 could be apply to mantra also

- 4*4 if you have still frame and then you go higher in max samples

- 8*8 for moving stills and then you will limit your max samples

- 12*12 if you have fur / dof / hard motion blur

Well it's an oversimplification but i think it's pretty true

 

I also note in most engine i have test that pushing the light quality pretty high and keeping pixel sampling as low as possible is a good practice.

 

Nice tips for the blurred and low res hdr i forgot to do that ..

 

 

I would be curious to get your pov about this. Few month ago i have done a very short prototype as a request feature for maxwell plugin.

It haven't been use at the end. So i would be curious to know if this kind a BSDF layering logic looks cool to you , or if you prefer the UBER monolithic path.

 

The shader is an empty box it do nothing, but i would be curious to get your POV on this because i could try to work on a mantra implementation.

Of course some maxwell exotic parameter could be rename. But do you like this logic of infinite BSDF combination ? or do you prefer the mantra shader logic ?

 

Cheers

 

E

 

 

BSDF_layering_shader.hip

Edited by sebkaine
Link to comment
Share on other sites

Thanks Jonathan ! i've already see that , but just a prototype for now ... :)

 

Ok i think we got ourself a bingo !  Thanks to all your advise i have reach my goal ...

 

Mantra 1080P => 26 minutesmantra_027_pbr_photon_26m42s.jpg

 

Settings :

- Env + Portal + Photon Map

- sample : 4*4

- min sample : 2

- max sample : 10

- noise : 0.004

- reflect limit : 2

- diffuse limit : 2

- min reflect ratio : 1

- light quality : 4

- photon filter : 1

- photon threshold : 1

 

I ask forgiveness for saying bad thing about mantra photon map ... this is not mental ray here ...

 

Mantra simply kick ass, i think i can even go under 20 minutes with :

- optimised VEX shader

- using the indirect sampling feature to optimise direct / indirect sample

 

Next time a lighter will said that Mantra suck or is slow i'm gonna have some few materials to discuss about ....

Edited by sebkaine
  • Like 3
Link to comment
Share on other sites

as a side note unlike V-Ray lightcache which is camera dependant (if my memory is correct).

it looks that photon map in mantra are not connected to the camera pov.

so if i am correct you can bake GI in a photon map for static asset and render a flythrough at no extra cost.

Link to comment
Share on other sites

I knew that it could be more optimized but you really went further, I got a 30 min render in 1080 with pixel samples 4X4 I am even curious to discover how you did you really achieve that last render time @sebkaine!

 

Awesome Job!

 

And yeah, Mantra is awesome!  :D

Link to comment
Share on other sites

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.

Guest
Reply to this topic...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

×
×
  • Create New...