Neon Sign - Part 2 - Rendering
14731 17 7- Keith Johnson
- Member
- 30 posts
- Joined: Oct. 2013
- Offline
- SreckoM
- Member
- 379 posts
- Joined: Dec. 2006
- Offline
- anon_user_37409885
- Member
- 4189 posts
- Joined: June 2012
- Offline
- Doudini
- Member
- 333 posts
- Joined: Oct. 2012
- Offline
- Werner Ziemerink
- Member
- 1268 posts
- Joined: March 2014
- Offline
- Keith Johnson
- Member
- 30 posts
- Joined: Oct. 2013
- Offline
- Keith Johnson
- Member
- 30 posts
- Joined: Oct. 2013
- Offline
There was a question on the vimeo comment board about render settings, so I thought I'd elaborate a bit here.
Renderer was PBR, and after tweaking render parameters in a bracketing/wedge fashion, I found the two most important variables in achieving quality versus speed were the pixel samples and noise level. Those are the main numbers I tweaked to flip between a low-quality/fast preview render and the hi-quality final render. The other variable that's a bit hard to quantify, was that I had to increase the shading quality on a few key pieces of geometry (I did this on a per object basis - bumping it up from the default 1 to something around 3-4). As I state in the video, this was mainly to fix what looked like an odd dicing error on large n-gons that were utilizing gridded bump maps to simulate panel seams.
As for the other settings - I found a law of diminishing returns kick-in with the Max Ray samples. It appeared anything over a value of 6 did little to the render. Also, on the limits tab, it was important to keep the Reflect, Refract, and Diffuse numbers as low as possible. Any values above 2,2,and 1 respectively, achieved only a minor bump in quality, but a huge increase in render times. So that's basically the unscientific way I arrived at my render settings! Screenshots for both the low-quality and final-quality settings are attached.
I found with low-quality settings, interactively tweaking the shading and lighting was pretty good…at least good enough for me to move forward without getting frustrated. As I state in the video. With the final quality setting, rendering at 2000 x 2000, the rendertimes were generally around 8 hours. This is on a 2nd generation i7 machine, capable of turboboost above 3 GHz with 12 gb ram. One thing I did notice is that my machine was very rapidly using up ALL of the ram with no headroom, so I imagine it would be a good thing to upgrade that. Thankfully, Mantra does not just crash on you when you hit the ram ceiling, like V-Ray used to be notorious for. Still, 8 hours for a 2000x2000 render seems a little steep. And it still may have been too noisy for animation. This didn't bother me for a small test project, but it could be an issue later if I do full animation work for a production.
Really though, the speed was my only gripe. Mantra was incredibly stable, reliable, and predictable. I don't think I had a single crash or freeze while working with the interactive render, and the dicing error I stated above was the only unexpected surprise that required troubleshooting.
Renderer was PBR, and after tweaking render parameters in a bracketing/wedge fashion, I found the two most important variables in achieving quality versus speed were the pixel samples and noise level. Those are the main numbers I tweaked to flip between a low-quality/fast preview render and the hi-quality final render. The other variable that's a bit hard to quantify, was that I had to increase the shading quality on a few key pieces of geometry (I did this on a per object basis - bumping it up from the default 1 to something around 3-4). As I state in the video, this was mainly to fix what looked like an odd dicing error on large n-gons that were utilizing gridded bump maps to simulate panel seams.
As for the other settings - I found a law of diminishing returns kick-in with the Max Ray samples. It appeared anything over a value of 6 did little to the render. Also, on the limits tab, it was important to keep the Reflect, Refract, and Diffuse numbers as low as possible. Any values above 2,2,and 1 respectively, achieved only a minor bump in quality, but a huge increase in render times. So that's basically the unscientific way I arrived at my render settings! Screenshots for both the low-quality and final-quality settings are attached.
I found with low-quality settings, interactively tweaking the shading and lighting was pretty good…at least good enough for me to move forward without getting frustrated. As I state in the video. With the final quality setting, rendering at 2000 x 2000, the rendertimes were generally around 8 hours. This is on a 2nd generation i7 machine, capable of turboboost above 3 GHz with 12 gb ram. One thing I did notice is that my machine was very rapidly using up ALL of the ram with no headroom, so I imagine it would be a good thing to upgrade that. Thankfully, Mantra does not just crash on you when you hit the ram ceiling, like V-Ray used to be notorious for. Still, 8 hours for a 2000x2000 render seems a little steep. And it still may have been too noisy for animation. This didn't bother me for a small test project, but it could be an issue later if I do full animation work for a production.
Really though, the speed was my only gripe. Mantra was incredibly stable, reliable, and predictable. I don't think I had a single crash or freeze while working with the interactive render, and the dicing error I stated above was the only unexpected surprise that required troubleshooting.
- mandrake0
- Member
- 644 posts
- Joined: June 2006
- Offline
in these post there are some tips for the rendering speed up's the 9 hours i find are very long:
http://forums.odforce.net/topic/16507-corona-render/page-2#entry101045 [forums.odforce.net]
http://forums.odforce.net/topic/22080-suppress-small-artefact-in-pbr-render/ [forums.odforce.net]
btw. a absolutly nice job you have done… :-)
http://forums.odforce.net/topic/16507-corona-render/page-2#entry101045 [forums.odforce.net]
http://forums.odforce.net/topic/22080-suppress-small-artefact-in-pbr-render/ [forums.odforce.net]
btw. a absolutly nice job you have done… :-)
- sl0throp
- Member
- 258 posts
- Joined:
- Offline
EDIT - renders are looking really good…
I think your settings can be adjusted. I think your noise settings are too extreme and your min and max ray samples too low, and I think you can probably turn stochastic transparency off.
I can still see noise under your neon lights that I think will clear up with the min max ray samples. For final quality I often start right around here
Pixel samples = 3 X 3 to 5 X5 - tune up but start at 3 X 3
Min Ray samples = 2
Max Ray samples = 12 ( Sometime it goes up to 18)
Noise level = .02 (I have seen in a post that this is less then the grain we would get from standard film)
I think that also…
You can increase your diffuse bounces to 2 or 3 - I don't see much of a time hit here in my testing, some but not much
Color limit = 5
Mantra is slow, but not that slow. I have certainly done production rendering at even larger sizes much faster than that.
Hopefully Redshift which I use in Maya will be available for Houdini soon, as that renderer is super nice and wicked fast…
I think your settings can be adjusted. I think your noise settings are too extreme and your min and max ray samples too low, and I think you can probably turn stochastic transparency off.
I can still see noise under your neon lights that I think will clear up with the min max ray samples. For final quality I often start right around here
Pixel samples = 3 X 3 to 5 X5 - tune up but start at 3 X 3
Min Ray samples = 2
Max Ray samples = 12 ( Sometime it goes up to 18)
Noise level = .02 (I have seen in a post that this is less then the grain we would get from standard film)
I think that also…
You can increase your diffuse bounces to 2 or 3 - I don't see much of a time hit here in my testing, some but not much
Color limit = 5
Mantra is slow, but not that slow. I have certainly done production rendering at even larger sizes much faster than that.
Hopefully Redshift which I use in Maya will be available for Houdini soon, as that renderer is super nice and wicked fast…
- Keith Johnson
- Member
- 30 posts
- Joined: Oct. 2013
- Offline
- Tobias Hoffmann
- Member
- 14 posts
- Joined: July 2014
- Offline
- Timm Wagener
- Member
- 7 posts
- Joined: June 2014
- Offline
- mzigaib
- Member
- 976 posts
- Joined: April 2008
- Offline
- verysame
- Member
- 59 posts
- Joined: May 2015
- Offline
There's a copy on YouTube:
https://www.youtube.com/watch?v=BpdO6vjU_24 [youtube.com]
https://www.youtube.com/watch?v=r0o09mpUnvo [youtube.com]
https://www.youtube.com/watch?v=BpdO6vjU_24 [youtube.com]
https://www.youtube.com/watch?v=r0o09mpUnvo [youtube.com]
Andrew
Houdini Apprentice
Houdini Apprentice
- mzigaib
- Member
- 976 posts
- Joined: April 2008
- Offline
Thanks for the links, there is also a 3rd part 3rd part [youtube.com]
But this is just the modeling part it is missing the render part, which is really a shame it is a awesome result.
But this is just the modeling part it is missing the render part, which is really a shame it is a awesome result.
- verysame
- Member
- 59 posts
- Joined: May 2015
- Offline
mzigaib
Thanks for the links, there is also a 3rd part 3rd part [youtube.com]
But this is just the modeling part it is missing the render part, which is really a shame it is a awesome result.
You're right, didn't notice there was the render part too.
It seems all the links around are pointing to the same Vimeo account.
Too bad there isn't a copy of those on YouTube.
Andrew
Houdini Apprentice
Houdini Apprentice
- Nicolas Heluani
- Member
- 280 posts
- Joined: Dec. 2015
- Offline
- Toby Williams-Ellis2
- Member
- 8 posts
- Joined: Dec. 2014
- Offline
Hi Keith,
I got loads out of your tutorials and was really interested in the one that you did with the pendulum wave, which I always meant to go back to look at at a later date, but then discovered had disappeared. Is this something you might repost at some point? I really enjoyed the way you'd explain your whole process beginning to end, including the source of your inspiration. Thanks for putting them up in the first place, in any case
I got loads out of your tutorials and was really interested in the one that you did with the pendulum wave, which I always meant to go back to look at at a later date, but then discovered had disappeared. Is this something you might repost at some point? I really enjoyed the way you'd explain your whole process beginning to end, including the source of your inspiration. Thanks for putting them up in the first place, in any case
-
- Quick Links