why mantra renderer is so so so slow? :? that ruins houdini :x
mental ray in other packages is a lot faster 8)
why mantra renderer is so so so slow?
71284 34 3- dyei
- Member
- 28 posts
- Joined: July 2009
- Offline
- evgen
- Member
- 63 posts
- Joined: Jan. 2008
- Offline
- sanostol
- Member
- 577 posts
- Joined: Nov. 2005
- Offline
- Stalkerx777
- Member
- 183 posts
- Joined: Nov. 2008
- Offline
dyei
why mantra renderer is so so so slow? :? that ruins houdini :x
mental ray in other packages is a lot faster 8)
Another one. Take a look at this thread [sidefx.com]. Try to answer some questions on the last page.
mentalray + displacement + motionblur = some trouble+ DOF == really trouble
Aleksei Rusev
Sr. Graphics Tools Engineer @ Nvidia
Sr. Graphics Tools Engineer @ Nvidia
- old_school
- Staff
- 2540 posts
- Joined: July 2005
- Offline
dyei
why mantra renderer is so so so slow? :? that ruins houdini :x
mental ray in other packages is a lot faster 8)
With scant details about your issue with Mantra performance, why should we even bother? Because we are so nice.
Please focus on the issue of performance. I don't mind number comparisons but your conclusion with absolutely no information leaves us all guessing about a lot of things. Compare MR with Mantra's various render engines on a specific scene and let's see what can be done. There are threads in the Side Effects forums that have comparisons. Most everyone that has done serious evaluations of Mantra did not come up with your conclusion that Mantra is slow in all cases, in all scenes, with all materials, with motion blur, with depth of field, with mp rendering only, with hybrid mp/rt, with rt, and with pbr, in simple scenes, in large scenes, utilizing extra image planes (aov's), in managing and executing render passes, in rendering fur, rendering point clouds, rendering volume primitives, utilizing mantra procedurals such as the new volume procedural?
Wow you really must have done some exhaustive testing! Please do share.
Mantra has several render engines. You don't mention anything.
What shaders do you have assigned. Again you don't mention anything.
Are you using environment based lighting? Yet again no mention.
Are you doing realistic rendering, npr, what? And again no mention.
No image to post of your progress….
Without any information at all, how can anyone duplicate your results to come to the same conclusion? Seriously?
As for indirect illumination which the previous posters somehow assumed was your issue (I don't know how btw as you gave us no details), there are a few ways to do that with Mantra as well. You can use the Indirect Light to generate indirect photon caches to accelerate this part of the render process.
Btw GI is just one of a few strategies to deal with indirect illumination (I am really starting to loathe mis-terminology these days…). I also have never heard of “smooth global illumination”? I've heard of noise in indirect lighting calculations and trying to reduce said noise requires increased samples which means increased render times in all physically based renderers including Arnold and Maxwell (both excellent physically based render engines) and Mantra pbr.
There's at least one school like the old school!
- Korhon
- Member
- 334 posts
- Joined: July 2007
- Offline
Hehehe! Spot on Jeff!! 8)
I use mantra everyday in production, even with GI. I think the reason people get scared of mantras globalillumination is that they`r not used to dealing with full raytraced gi. When using mentalrays finalgather you will offcourse get smaler rendertimes. But you will also get flickering and poor quality, since it does a approximation of the lightbounces.
Mantra Pwns Mentalray bigtime in my opinion!
I use mantra everyday in production, even with GI. I think the reason people get scared of mantras globalillumination is that they`r not used to dealing with full raytraced gi. When using mentalrays finalgather you will offcourse get smaler rendertimes. But you will also get flickering and poor quality, since it does a approximation of the lightbounces.
Mantra Pwns Mentalray bigtime in my opinion!
- JasonSlab
- Member
- 1535 posts
- Joined: March 2020
- Offline
yes, when it comes to full blown production, mantra kicks MR's ass!
i'd also like to see some comparisons?
jason
i'd also like to see some comparisons?
jason
fx and lighting @ rebelway
https://www.youtube.com/@jason_slab
https://www.youtube.com/@jason_slab
- evgen
- Member
- 63 posts
- Joined: Jan. 2008
- Offline
Well, Vray's Animated Irradiance map ideal solution for getting fast flicker free smooth result without noise. PBR is good for outdoor scenes, but indoor is big problem even with lightportals and photons. I made a few tests with mantra and must to say that mantra has a fastest direct pathtracing GI. http://files.evgen3d.com/test_PBR_photons_11mins_cor.jpg [files.evgen3d.com]
But without any “fake” solution. Irradiance cache doesn't give production ready result for animation and in some cases for static images too.
for attached image i cant get smooth result even with 6 time higher samples values and AA. Calculating of attached image tooks 11 mins. but if i want to get noise free image, even 1 hour render doesnt gives smooth image i attached scene for this image
But without any “fake” solution. Irradiance cache doesn't give production ready result for animation and in some cases for static images too.
for attached image i cant get smooth result even with 6 time higher samples values and AA. Calculating of attached image tooks 11 mins. but if i want to get noise free image, even 1 hour render doesnt gives smooth image i attached scene for this image
- Soothsayer
- Member
- 874 posts
- Joined: Oct. 2008
- Offline
- kuba
- Member
- 345 posts
- Joined:
- Offline
- old_school
- Staff
- 2540 posts
- Joined: July 2005
- Offline
Watch Andrew Clinton's talk on Mantra. Near the end he describes what the noise parameter does and how it works with the min and max sample parameters.
I do what Andrew mentions in the video. Increase primary samples then raise the max samples up and use noise to increase the max samples. Noise I believe is the difference in color value of neighboring pixels. Anything above the noise threshold (i.e. the change in neighboring color values) is considered “noise” and will therefore cause more samples to be generated in those areas up to the max sample value to reduce the noise.
I do what Andrew mentions in the video. Increase primary samples then raise the max samples up and use noise to increase the max samples. Noise I believe is the difference in color value of neighboring pixels. Anything above the noise threshold (i.e. the change in neighboring color values) is considered “noise” and will therefore cause more samples to be generated in those areas up to the max sample value to reduce the noise.
There's at least one school like the old school!
- evgen
- Member
- 63 posts
- Joined: Jan. 2008
- Offline
- jason_iversen
- Member
- 12672 posts
- Joined: July 2005
- Offline
I see you've already set your Diffuse and Reflection Limits way down – but I see your sampling settings aren't really ideal.
So: the I've found the relying noise variance algorithm to provide you clean illumination is not ideal. We've just done 200 shots using PBR and these are the sampling settings that worked best for us:
- Pixel Samples: 7x7 (up to 10x10 if the scene really needed it)
- Min/Max Ray Samples: 1/9 – sometimes 2/9
- Noise Level: 0.08 (this seemed good enough to us … less than film grain)
- Reflect/Diffuse Limits: 2, 2
- Color Space: Gamma 2.2
- Color Limit: 4 (your 1 is clamping way too hard)
- Photons: 2m photons for a huge room. Yours is 7m! I'm sure 1m is fine for this space. Ratio was 2.5.
Be very careful of displacement. The default Mantra Surface shader is less efficient than a Plastic shader because of mere presence of the displacement shader. Our shader manages this with the “Disable Displace Shader Rendering” property but the distributed Mantra Surface-based shaders don't.
So: the I've found the relying noise variance algorithm to provide you clean illumination is not ideal. We've just done 200 shots using PBR and these are the sampling settings that worked best for us:
- Pixel Samples: 7x7 (up to 10x10 if the scene really needed it)
- Min/Max Ray Samples: 1/9 – sometimes 2/9
- Noise Level: 0.08 (this seemed good enough to us … less than film grain)
- Reflect/Diffuse Limits: 2, 2
- Color Space: Gamma 2.2
- Color Limit: 4 (your 1 is clamping way too hard)
- Photons: 2m photons for a huge room. Yours is 7m! I'm sure 1m is fine for this space. Ratio was 2.5.
Be very careful of displacement. The default Mantra Surface shader is less efficient than a Plastic shader because of mere presence of the displacement shader. Our shader manages this with the “Disable Displace Shader Rendering” property but the distributed Mantra Surface-based shaders don't.
Jason Iversen, Technology Supervisor & FX Pipeline/R+D Lead @ Weta FX
also, http://www.odforce.net [www.odforce.net]
also, http://www.odforce.net [www.odforce.net]
- uniqueloginname
- Member
- 330 posts
- Joined: July 2005
- Offline
- Soothsayer
- Member
- 874 posts
- Joined: Oct. 2008
- Offline
With the little experience that I have this is what I found. My thumb rules:
-Pixel samples have an effect on aliasing, motion blur and DOF “noise”. Image details react to it.
-Ray Samples have an effect on “GI noise”. For this setting the Noise Level is the most important factor in both quality and render time.
-Reflect/refract llimits hardly ever impact quality/render time. Diffuse limit does though.
-Shading dicing, even a little change, can help reduce noise. Very expensive!
-Ray-Z can be crucial for displacement quality.
-PBR is almost always superior in quality and time.
-With Micropolygon, correct Tile Size is very important!
-Pixel samples have an effect on aliasing, motion blur and DOF “noise”. Image details react to it.
-Ray Samples have an effect on “GI noise”. For this setting the Noise Level is the most important factor in both quality and render time.
-Reflect/refract llimits hardly ever impact quality/render time. Diffuse limit does though.
-Shading dicing, even a little change, can help reduce noise. Very expensive!
-Ray-Z can be crucial for displacement quality.
-PBR is almost always superior in quality and time.
-With Micropolygon, correct Tile Size is very important!
--
Jobless
Jobless
- JasonSlab
- Member
- 1535 posts
- Joined: March 2020
- Offline
jason_iversen
I see you've already set your Diffuse and Reflection Limits way down – but I see your sampling settings aren't really ideal.
So: the I've found the relying noise variance algorithm to provide you clean illumination is not ideal. We've just done 200 shots using PBR and these are the sampling settings that worked best for us:
- Pixel Samples: 7x7 (up to 10x10 if the scene really needed it)
- Min/Max Ray Samples: 1/9 – sometimes 2/9
- Noise Level: 0.08 (this seemed good enough to us … less than film grain)
- Reflect/Diffuse Limits: 2, 2
- Color Space: Gamma 2.2
- Color Limit: 4 (your 1 is clamping way too hard)
- Photons: 2m photons for a huge room. Yours is 7m! I'm sure 1m is fine for this space. Ratio was 2.5.
interesting settings Jason, are the shot you guys rendering mainly diffused surfaces?
we're busy with some shots at the moment that have loads of reflections, i'm stil getting really noisy images with
-pixel samples 7x7
-Min/Max 6/16
my main problem is the environment is quite dark, bright lights with quick falloffs
jason
fx and lighting @ rebelway
https://www.youtube.com/@jason_slab
https://www.youtube.com/@jason_slab
- Stalkerx777
- Member
- 183 posts
- Joined: Nov. 2008
- Offline
jason_iversen
I see you've already set your Diffuse and Reflection Limits way down – but I see your sampling settings aren't really ideal.
So: the I've found the relying noise variance algorithm to provide you clean illumination is not ideal. We've just done 200 shots using PBR and these are the sampling settings that worked best for us:
- Pixel Samples: 7x7 (up to 10x10 if the scene really needed it)
- Min/Max Ray Samples: 1/9 – sometimes 2/9
- Noise Level: 0.08 (this seemed good enough to us … less than film grain)
- Reflect/Diffuse Limits: 2, 2
- Color Space: Gamma 2.2
- Color Limit: 4 (your 1 is clamping way too hard)
- Photons: 2m photons for a huge room. Yours is 7m! I'm sure 1m is fine for this space. Ratio was 2.5.
Be very careful of displacement. The default Mantra Surface shader is less efficient than a Plastic shader because of mere presence of the displacement shader. Our shader manages this with the “Disable Displace Shader Rendering” property but the distributed Mantra Surface-based shaders don't.
Thx for sharing info jason. But i'm really wonder, how can you get low-noise results, with this, pretty low sampling settings?
Aleksei Rusev
Sr. Graphics Tools Engineer @ Nvidia
Sr. Graphics Tools Engineer @ Nvidia
- jason_iversen
- Member
- 12672 posts
- Joined: July 2005
- Offline
We are rendering every kind of surface: tiles, metals, glass, enamel, stone, wood – I'd post an image if I was allowed to.
@jasonslab, I assume your Noise Limit is low, and you're using Gamma2.2? We found that it's much better to use Pixel Samples than boost min/max ray samples – it improves everything: edges, motion-blur, illumination. If you have a Min Ray > 2, you may as well rather bump Pixel Samples and bring Min Ray down.
What is your environment? Most of ours is a large room will with machinery – and we render off panoramic renders for the ceiling and walls and use those as envlights – and this really helps since it can be re-used, and rays can escape the scene. This is the type of workflow optimization you need to consider to make it practical.
@stalker, What helps is that our environment is realistically surfaced. It has texture, bump, scratches, dirt, etc. It's not an artificially clinical test scene like everyone tests with – that might only be used in 10% of car commercials. The texture masks some of the noise which makes it OK. Not that I don't think Mantra could be sped up and improved! If it can render pristine “clinical” environments nice and fast, it can render highly texured scenes even faster. Mantra does need improvement, no doubt at all!
@jasonslab, I assume your Noise Limit is low, and you're using Gamma2.2? We found that it's much better to use Pixel Samples than boost min/max ray samples – it improves everything: edges, motion-blur, illumination. If you have a Min Ray > 2, you may as well rather bump Pixel Samples and bring Min Ray down.
What is your environment? Most of ours is a large room will with machinery – and we render off panoramic renders for the ceiling and walls and use those as envlights – and this really helps since it can be re-used, and rays can escape the scene. This is the type of workflow optimization you need to consider to make it practical.
@stalker, What helps is that our environment is realistically surfaced. It has texture, bump, scratches, dirt, etc. It's not an artificially clinical test scene like everyone tests with – that might only be used in 10% of car commercials. The texture masks some of the noise which makes it OK. Not that I don't think Mantra could be sped up and improved! If it can render pristine “clinical” environments nice and fast, it can render highly texured scenes even faster. Mantra does need improvement, no doubt at all!
Jason Iversen, Technology Supervisor & FX Pipeline/R+D Lead @ Weta FX
also, http://www.odforce.net [www.odforce.net]
also, http://www.odforce.net [www.odforce.net]
- FXGuru
- Member
- 2 posts
- Joined: Feb. 2011
- Offline
Going to try Jason's settings. Fingers crossed :wink:
Thanks.
——————————————————-
www.3drender.co.uk - Pro-Viz(TM) 3D Textures
Thanks.
——————————————————-
www.3drender.co.uk - Pro-Viz(TM) 3D Textures
- uniqueloginname
- Member
- 330 posts
- Joined: July 2005
- Offline
-
- Quick Links