Hi guys.
I decided to try some lighting practice with Houdini since version 16 is out. This build seems significantly faster than 15.5. Rendering this in 15.5 was taking nearly 2 hours whilst with 16 is taking 58 minutes per frame.
Still a long way to go but so far it's coming along nicely.
So info of the scene is pretty simple in lighting terms, 1 spotlight for the sun, 1 area light for filling the room and a GI light for GI/caching.
Mantra settings are:
Sampling: 3x3
Noise level: 0.005
Min/Max Samples: 1/64
Diffuse Limit: 1
Model created and supplied by: Georgio Luciano.
Interior Rendering
7789 10 1
-
- ragupasta
- Member
- 349 posts
- Joined: Aug. 2006
- Offline
-
- edward
- Member
- 7868 posts
- Joined: July 2005
- Offline
Have you considered using a portal light for the sun? eg tutorial: https://vimeo.com/127279992 [vimeo.com]
http://www.sidefx.com/docs/houdini/shelf/portallight [sidefx.com]
http://www.sidefx.com/docs/houdini/shelf/portallight [sidefx.com]
Edited by edward - March 2, 2017 22:14:06
-
- ragupasta
- Member
- 349 posts
- Joined: Aug. 2006
- Offline
Thanks for the advice and link Edward.
So I gave the portal light method a try. I have to keep the GI Light in the scene otherwise the portal light wont get enough light in, even with a very high intensity setting. Rendertimes seem very similar to the age old method I was using of area lights. I dropped the max ray samples to 32 as a test, and the rendertime came out at 40:14. With the max samples back to 64 I can easily see another 10 minutes rendertime added, so performance wise both methods seem similar.
Obviously noise has returned due to lower sampling value for the dark areas.
Quick question Edward if you can answer it.
H15 had a parameter on the Mantra node called “Indirect ray samples” for directly controlling GI noise area's. Having a look around at Mantra in H16, I am assuming this is now controlled by the “Diffuse Quality” slider under the sampling tab.
So I gave the portal light method a try. I have to keep the GI Light in the scene otherwise the portal light wont get enough light in, even with a very high intensity setting. Rendertimes seem very similar to the age old method I was using of area lights. I dropped the max ray samples to 32 as a test, and the rendertime came out at 40:14. With the max samples back to 64 I can easily see another 10 minutes rendertime added, so performance wise both methods seem similar.
Obviously noise has returned due to lower sampling value for the dark areas.
Quick question Edward if you can answer it.
H15 had a parameter on the Mantra node called “Indirect ray samples” for directly controlling GI noise area's. Having a look around at Mantra in H16, I am assuming this is now controlled by the “Diffuse Quality” slider under the sampling tab.
Edited by ragupasta - March 4, 2017 06:02:17
-
- mark
- Staff
- 2619 posts
- Joined: July 2005
- Offline
ragupasta
Quick question Edward if you can answer it.
H15 had a parameter on the Mantra node called “Indirect ray samples” for directly controlling GI noise area's. Having a look around at Mantra in H16, I am assuming this is now controlled by the “Diffuse Quality” slider under the sampling tab.
Hi… There was an output plane toggle to inspect indirect ray samples called “Indirect ray samples” – and I think that's still there in H16.0.
In H15.5 there were parameters “Min Indirect Ray Samples” and “Max Indirect Ray Samples”, along with “Indirect Noise Level”. These parameters also exist in H16.0.
The Mantra User Guide has a good discussion about the meaning of Diffuse Quality in the section on “Indirect Illumination”.
http://www.sidefx.com/docs/houdini/render/noise [sidefx.com]
P.S. This part of the documentation hasn't been updated for H16.0 yet, but it should give a reasonable indication of what's going on.
Edited by mark - March 5, 2017 16:14:18
-
- ragupasta
- Member
- 349 posts
- Joined: Aug. 2006
- Offline
-
- ragupasta
- Member
- 349 posts
- Joined: Aug. 2006
- Offline
So I am having another go with mantra with a more geometrically heavy scene. You have all seen this scene before due to the lighting challenges over at CGTalk. Files are hosted at Jeremy Birns site over at www.3drender.com
From a Mantra perspective the settings are pretty much the same as the last scene, but lighting consists of:
1x spotlight for the sun
4x portal lights for room fill
1x GI light for caching
Basic lighting and no scene shaders as of yet. Rendetime stands at:- 34:55 which is very much acceptable in my eyes.
The more I'm playing with lighting/mantra the more I'm liking it.
The sunlight is a tad too strong at the moment, but lets see after I build some shaders.
Modeled by Alvaro Luna Bautista and Joel Andersdon.
From a Mantra perspective the settings are pretty much the same as the last scene, but lighting consists of:
1x spotlight for the sun
4x portal lights for room fill
1x GI light for caching
Basic lighting and no scene shaders as of yet. Rendetime stands at:- 34:55 which is very much acceptable in my eyes.
The more I'm playing with lighting/mantra the more I'm liking it.
The sunlight is a tad too strong at the moment, but lets see after I build some shaders.
Modeled by Alvaro Luna Bautista and Joel Andersdon.
-
- ragupasta
- Member
- 349 posts
- Joined: Aug. 2006
- Offline
Additional question: Atmospheric effects such as volume lighting (aka: god rays).
The old method of lit fog/atmospheric object is notoriously slow so that is not going to be an effective way of doing it.
The next logical solution would be to create a volume fog to encompass the scene and render it that way. Problem is when using say a billowy smoke shader, when dropping the density value very low to actually be able to get light through the density it also becomes very slow and noisy as well. This means significantly increasing the volume samples in Mantra and possibly the stochastic samples depending on the situation.
Would using VDB be a better solution? And if so what would be the best workflow?
VolumeSOP to encomass the scene then use a convertVDB set to VDB?
Not really played much with VDB's so it's an unfamiliar area for me.
Any suggestions?
The old method of lit fog/atmospheric object is notoriously slow so that is not going to be an effective way of doing it.
The next logical solution would be to create a volume fog to encompass the scene and render it that way. Problem is when using say a billowy smoke shader, when dropping the density value very low to actually be able to get light through the density it also becomes very slow and noisy as well. This means significantly increasing the volume samples in Mantra and possibly the stochastic samples depending on the situation.
Would using VDB be a better solution? And if so what would be the best workflow?
VolumeSOP to encomass the scene then use a convertVDB set to VDB?
Not really played much with VDB's so it's an unfamiliar area for me.
Any suggestions?
-
- tamte
- Member
- 8766 posts
- Joined: July 2007
- Offline
you should try just closed polygonal object like box around your scene
and assign Uniform Volume material from the gallery
it's more optimised for constant density fog and has separate control for sampling from mantra ones
and assign Uniform Volume material from the gallery
it's more optimised for constant density fog and has separate control for sampling from mantra ones
Edited by tamte - March 21, 2017 21:18:17
Tomas Slancik
FX Supervisor
Method Studios, NY
FX Supervisor
Method Studios, NY
-
- KaiStavginski
- Staff
- 654 posts
- Joined: Aug. 2013
- Offline
Here's a scene with a uniform volume, as suggested by tamte.
I agree that this should be the best approach. Make sure neither pixel samples nor the sampling on the uniform volume shader are too low. A good balance between the two is key.
This is with 10x10 pixel samples and 128 volume samples on the shader. Without bounces or GI light though.
It took 23 minutes on 8 cores.
I agree that this should be the best approach. Make sure neither pixel samples nor the sampling on the uniform volume shader are too low. A good balance between the two is key.
This is with 10x10 pixel samples and 128 volume samples on the shader. Without bounces or GI light though.
It took 23 minutes on 8 cores.
Edited by KaiStavginski - March 23, 2017 06:44:42
Kai Stavginski
Senior Technical Director
SideFX
Senior Technical Director
SideFX
-
- chevita
- Member
- 122 posts
- Joined: Aug. 2013
- Offline
-
- Mirko Jankovic
- Member
- 162 posts
- Joined: Aug. 2015
- Online
-
- Quick Links