Keep the Camera AA samples at a minimum (3 or 4) and switch the adaptive sampling on. Then set the max AA samples to something like 30 or 50 and set the adaptive threshold to something in between 0.02 and 0.01 (this way you don't force the engine to get clamped by the Camera AA and you tweak the quality via the threshold). I would say that my main complaints with this beta are the following: 1) the need to "Pre-Populate the GPU caches" (it's recommended) before even starting the render, which can take up to 15 minutes (yes, fifteen minutes). Thanks God it's just one off and you don't have to do it again, and on top of that it makes the next renders start without delay. 2) No support for Area Lights. 3) No support for "Closest" filters, blockers or decay. 4) No support for light filters. 5) No support for shadow matte. 6) Selecting a specific GPU (I have three installed) crashes the software (so I have to leave all my GPUs active for the renderer). A lot of work to do and a long way to go for this feature to be production proof. I still find the interactivity way way way faster on CPU than GPU (but render times are of course faster on the GPU).
Perhaps being into the habit of getting quick feedback, but it has not sunk in yet that the quick feedback happens realtime with the GPU as he was demonstrating
Will definitely stick to Redshift until Arnold GPU is mature. If the only point is the possibility to be more cost-effective with a CPU farm, I think it doesn't make sense. Just give it some time and the GPU farm will get cheaper. The other thing I am concern about is the development for c4d/Houdini knowing Arnold is an Autodesk product. I think it's a bit more healthy to have a more independent render engine. One question Chad about the switches you have in your layout for changing the viewport background: can you tell me who's doing it? Heard of it before but can't find it online anymore. Many thanks.
I think there an option that could speed up scene loading time, its called Pre-populate GPU Cache, u can found it in the utilities, give it a try maybe it will work
I'm trying to use my gpu to render quicker. As Arnold does not support volumes on GPU is there a way to use AOV's to render fumefx on cpu and GPU for everything else?
I think your spot on with the indirect light issues, I made a box in Maya and shined a directional light in it and the result wasn't great. I also couldn't get the portal light to work while in GPU mode.
It's good not only for cloud based rendering but as a small render farm owner with half my power in gpu and half in cpu I'm greatly happy to be able to use 100% of my farm power!!!!
Hi There! :) Can I ask these question? 1.) I want to be a 2d motion graphic designer, do I need to learn 3D and Cinema 4D 2.) Can I use Blender for 3D animation 3.) Why Cinema 4D is the industry standard for Motion Graphics? Why not Maya or Blender?
1) Not necessarily, but having some 3D knowledge will help a lot. For example, how to work in a 3D space. 2) Yes you can, all 3D programs are capable of modeling, rigging and animating. I would advise learning Maya if you are passionate enough about 3D animation, Blender is a good start, once you realize what type of 3D animation you want to pursue, you will later on either choose Maya, Cinema 4D or Houdini once you become really good in 3D. 3) The MoGraph system. Cinema 4D is THE choice for 3D Motion Graphics. Maya has MASH, but it still isnt good enough. And Houdini... Well, in Houdini, you literally create the tools yourself. But if your main interest is 3D motion graphics, then your main programs should be Cinema 4D and After Effects. So get going and start learning!
@@OGHearty I am contented for the feature of Blender especially the 2.8 version but Still Cinema 4d is considered as Motion Graphics Best Friend and still the standard when it comes to MoGraph BTW Thanks for answering me! :)
I love this video. Very honest! I wouldn't use Arnold GPU for rendering! what is the point of use GPU if you going to spend more time rendering and still having noise and issues!? Arnold should cheek this video.
Wouldn't put a minimum of 1 samples with adaptive turned on help on those large white areas on that final scene? I would also like to see that final scene with the denoiser turned on for the IPR.
is it possible to render large size scenes with gpu? most of the time when i tried to do it i got error like 'not enough memory'. i think it's the scene size should not exceed gpu memory.
This is my issue, I'm running 1080's in sli and still getting that error from a single character render. Prob about 40 million polys after extdsp though
Has anyone else run into problems with Fireflies using the Physical Sun/Sky? I have the options checked on, but can't get Direct and Indirect clamping working.
I completely regret switching to an AMD Radeon card last year, not because it's a bad card, but because of all the GPU rendering I would be missing out; particularly the new Arnold GPU renderer. I will be switching back to nVidia eventually, rest assured.
I've seen it on another GSG video I think and would also love to know where they got it. I guess for now, since we don't have something as cool as that head and all of its PBR maps to play with in octane, we will have to take it for granted that Arnold/Arnold GPU is magic.
No. All major GPU renderers require CUDA, and hence run only on Nvidia cards. The only exception to this is Prorender, which does run on AMD cards but is kind of a joke right now.
Where do you get the money to buy all these super-expensive render programs (you seem to have them all) other plug-ins, and on top of that the full version of C4D AND a beefy machine!? It must be atleast $20,000
GPU time to first pixel is NEVER going to be as fast as it is on the CPU. CPU rendering in Arnold already blows away most GPU renders in terms of interactivity.
Do you know the technical reason why out of interest? Naively I would have thought the GPU would be faster as in games they render hundreds of frames per second with virtually no lag
@@BENBOBBY My two cents: Firstly you'll have to load everything from disk into ram and then again have specialized a structure to pack data and send to the GPU which takes up some more time. You would have to do this everytime you change something in the scene. Games are faster because they use a different technique called rasterization. With ray tracing or path tracing there is an issue of thread divergence where more scattered the rays are the slower the performance. GPU threads run in lock step and if there is scattered data access or different parts of the code executing for different rays then you would see really bad performance
most renderers now a days tend to use one control for their glossy,reflection,AA,motion blur/dof,GI etc. at first I was not happy that I cannot control one aspect of it but the reason for it is that it will became much faster when your test and for final settings just change one setting even corona and vray works like this... this method is the reason why people like corona because you can control your render quality with just the noise value. even Clarisse' path tracing renderer works this way just one parameter to control the noise quality.
waiting 10 seconds is a drag!? boy you know nothing of gpu rendering. 10 seconds time to transfer data to the gpus is very Very fast. As a long time Octane render user trust me i know. Its one thing dealing with small scenes its another when scenes gets gig's heavy. though the rendering itself may be fast often the updating of info and data can severly bottle neck GPU workflow. because all the gpus have to be updated then re-synced. so again 10 secs is nothing be thankful.. =
Not impressed, my 5820k is faster then 2 980 ti`s. Takes forever to get a clean looking image when using Arnold gpu, guess is just for lookdev for now.
@@cgpov they cant Chad unless they write their own GPU implementation from the ground up. Something a lot of people are not talking about (we discussed this internally during beta) was the issue with Optix API for ray-tracing does not allow for local sampling. It cant do more samples on GI/Spec/etc. This is why it's forced to run as a pure path-tracer randomly firing only primary rays. They either need to find some crazy workaround or Nvidia needs to further develop the Optix API to allow for local sampling. Both are a very tall order.
@@TheArtofSaul Thanks! Yah it is a bit disappointing for sure. I'd be stoked if Redshift released 3.0 and it magically fixed the issues I have. I'd also love to see RS improve its UX overall. There are so many small annoying things that add up in RS.
Chad Ashley Big things got announced yesterday at GTC, random walk sss, multiscatter volumes, automatic sampling, much higher trace depths, shader sdk, and much more. The big one is the new real-time Redshift RT mode, basically like unreal engine or blenders eevee on steroids for viewports and even renders. Hopefully Nvidia puts up the talk soon. 3.0 initial builds are supposed to drop next month. (April) , just put in those wish list requests in the forums and the devs are very adamant about fixing things and adding features.
The gpu rendering still suck till today ... uses to much memory for such a simple scene with 4k texture ... i switch to octane and its not using nearly half of my 6gb gpu memory ...
If you are using R19 and want to avoid problems and waste of time just stay away from arnold, the last version I have used was so buggy, the 2.5.0 seems better but still I have a lot of problems of crashes, preview of materials and the GPU with ipr doesn't show anything after 1 min of calculations for a 1 million polygons scene , not very impressive... With R20 however, it's better and GPU seems to work correctly but still I get crashes too so I won't recommand arnold as a main renderer, vray, redshift, thea and octane are far better
I love Arnold and have been using it for years. I dont have a big GPU . I actually need a CPU renderer. I guess what you are saying is, if you want to use a GPU renderer, you wont recommend Arnold. Or are you saying stay away from Arnold regardless because I wouldnt agree with that. its been solid for me for years and still is. (Cpu)
These scenes are so light not really production stuff mental ray would be just as interactive 2years ago with that Geometry and deferred render setup, buy Xeons,
Did you mention what gpu is in your system? Changing the number of samples and activating or deactivating adaptive sampling between cpu and gpu isn't very usefull to convey how both methods compare. You didn't show the difference between previous randomwalk SSS and this new version. Sorry but this was a pretty bad first impressions video, I still don't have a clear idea about this gpu beta aside from the long time to load the scene. 👎
Thank Chad for all the tut about Arnold. Keep going my friend.
You're welcome! 👍
I'm all for it. Quality way better than OC or RS.
Thumbnail is perfect for a good night sleep.
Keep the Camera AA samples at a minimum (3 or 4) and switch the adaptive sampling on. Then set the max AA samples to something like 30 or 50 and set the adaptive threshold to something in between 0.02 and 0.01 (this way you don't force the engine to get clamped by the Camera AA and you tweak the quality via the threshold).
I would say that my main complaints with this beta are the following:
1) the need to "Pre-Populate the GPU caches" (it's recommended) before even starting the render, which can take up to 15 minutes (yes, fifteen minutes). Thanks God it's just one off and you don't have to do it again, and on top of that it makes the next renders start without delay.
2) No support for Area Lights.
3) No support for "Closest" filters, blockers or decay.
4) No support for light filters.
5) No support for shadow matte.
6) Selecting a specific GPU (I have three installed) crashes the software (so I have to leave all my GPUs active for the renderer).
A lot of work to do and a long way to go for this feature to be production proof.
I still find the interactivity way way way faster on CPU than GPU (but render times are of course faster on the GPU).
Thanks Chad! Great overview
You're welcome! 👍
Thank you for that presentation Chad !
You're very welcome
So waiting 10 seconds for a render to start is a problem and annoying, but waiting 3 times as long for it to finish on a cpu is not?
indeed, that made no sense at all.
Perhaps being into the habit of getting quick feedback, but it has not sunk in yet that the quick feedback happens realtime with the GPU as he was demonstrating
@@zimonslot he's comparing it to octane and redshift.
Are you going to add Arnold GPU to your workflow, or what render engine will you stick with?
What about corona renderer? It would be really nice, if you could make a video with some impressions about it
Yeah i was wandering why Chad was never interested in Corona render ???
or octane for that matter, which is also really powerful for mograph
Will definitely stick to Redshift until Arnold GPU is mature. If the only point is the possibility to be more cost-effective with a CPU farm, I think it doesn't make sense. Just give it some time and the GPU farm will get cheaper. The other thing I am concern about is the development for c4d/Houdini knowing Arnold is an Autodesk product. I think it's a bit more healthy to have a more independent render engine. One question Chad about the switches you have in your layout for changing the viewport background: can you tell me who's doing it? Heard of it before but can't find it online anymore. Many thanks.
@@kentwood06 yah those background gradient scripts are something I wrote (with the help of Jon Bosley)
I think there an option that could speed up scene loading time, its called Pre-populate GPU Cache, u can found it in the utilities, give it a try maybe it will work
I'm trying to use my gpu to render quicker. As Arnold does not support volumes on GPU is there a way to use AOV's to render fumefx on cpu and GPU for everything else?
I think your spot on with the indirect light issues, I made a box in Maya and shined a directional light in it and the result wasn't great. I also couldn't get the portal light to work while in GPU mode.
What GPU's are you running?
It's good not only for cloud based rendering but as a small render farm owner with half my power in gpu and half in cpu I'm greatly happy to be able to use 100% of my farm power!!!!
So cool. I don't have a need to have Arnold yet, but I love it!
Right on time Chad :)
i know I am kinda randomly asking but does anybody know a good place to stream new tv shows online ?
@Ashton Alfonso I use FlixZone. Just search on google for it =)
Be good if it could use the gpu AND cpu simultaneously! (I believe Thea Renderer can do this)
Cycles and v-ray also do that.
@@STROBdotNET Didn't know Cycles could do that too, interesting!
@@BENBOBBY th-cam.com/video/FJIEH9PsN38/w-d-xo.html
Not regarding this particular video. What is the point of using pure DOF instead of applying it in the post?
Hi Chad what is your final impressions on now Arnold 6 have been out?
Hi There! :)
Can I ask these question?
1.) I want to be a 2d motion graphic designer, do I need to learn 3D and Cinema 4D
2.) Can I use Blender for 3D animation
3.) Why Cinema 4D is the industry standard for Motion Graphics? Why not Maya or Blender?
1) Not necessarily, but having some 3D knowledge will help a lot. For example, how to work in a 3D space.
2) Yes you can, all 3D programs are capable of modeling, rigging and animating. I would advise learning Maya if you are passionate enough about 3D animation, Blender is a good start, once you realize what type of 3D animation you want to pursue, you will later on either choose Maya, Cinema 4D or Houdini once you become really good in 3D.
3) The MoGraph system. Cinema 4D is THE choice for 3D Motion Graphics. Maya has MASH, but it still isnt good enough. And Houdini... Well, in Houdini, you literally create the tools yourself. But if your main interest is 3D motion graphics, then your main programs should be Cinema 4D and After Effects. So get going and start learning!
@@OGHearty
I am contented for the feature of Blender especially the 2.8 version but Still Cinema 4d is considered as Motion Graphics Best Friend and still the standard when it comes to MoGraph
BTW Thanks for answering me! :)
Hmmm, intersting, what about Arnold GPU today? Sep 2021... ?
I love this video. Very honest! I wouldn't use Arnold GPU for rendering! what is the point of use GPU if you going to spend more time rendering and still having noise and issues!? Arnold should cheek this video.
How can i find model?
Wouldn't put a minimum of 1 samples with adaptive turned on help on those large white areas on that final scene? I would also like to see that final scene with the denoiser turned on for the IPR.
Min sample of 1 is not recommended. VERY noisy. The samples shown follow settings recommended by the Arnold team. 👍
The headscene probably renders fast because the head gets instanced. So the GPU only loads it once.
can you do another video for Cinema 4D Arnold 7 GPU render?
is it possible to render large size scenes with gpu?
most of the time when i tried to do it i got error like 'not enough memory'. i think it's the scene size should not exceed gpu memory.
AB186, which is render? It is must to work in renders which support "out from core". I think it's Redshift (and, possible Octane 4 (not 3) )
i think maybe it's a ram problem
This is my issue, I'm running 1080's in sli and still getting that error from a single character render. Prob about 40 million polys after extdsp though
Well, this is going to be great! Out of curiosity, Chad, what monitor do you use?
That's the LG 32GK850G-B.
@@Greyscalegorilla Thank you!
Has anyone else run into problems with Fireflies using the Physical Sun/Sky? I have the options checked on, but can't get Direct and Indirect clamping working.
Really cool Chad. Can we use a GPU Renderfarm frot he final sequence? I've heard that you can't render final with GPU.
You can render final frames in GPU, but it's still in beta so use at your own risk. Also there are many features that haven't been brought to GPU yet.
I completely regret switching to an AMD Radeon card last year, not because it's a bad card, but because of all the GPU rendering I would be missing out; particularly the new Arnold GPU renderer. I will be switching back to nVidia eventually, rest assured.
Hey, you can't record a video while chatting with your twin brother. You can not leave him talking alone! :.(
gosto muito dos seus videos, aprendo muito com eles..obrigado
do u have a transformer tutorial on how to make the transformer effect with anything
So to get the the "public beta" download the 30 day trail? Unfortunately i am unable to find the download for the beta.
www.arnoldrenderer.com/arnold/download/
the 2.5 version contains the 5.3 core.
Which render engineer is better to stick with?? Red shift or ..??
Well remember this is just a beta. So I'd stick with whatever works best for you. Luckily most of them have trials so you can test them all out
That thumbnail fking scated me!
Any news on AMD support in Arnold?
Are you try set low Adaptive threshold in GPU mode?
Yes, it made very little difference in my tests unfortunately.
Is that head available for download somewhere? Thanks for the vid!
I've seen it on another GSG video I think and would also love to know where they got it. I guess for now, since we don't have something as cool as that head and all of its PBR maps to play with in octane, we will have to take it for granted that Arnold/Arnold GPU is magic.
My GPU (AMD) is not showing up in the list, even with updated DRIVERS. Help? AMD is compatible?
No. All major GPU renderers require CUDA, and hence run only on Nvidia cards. The only exception to this is Prorender, which does run on AMD cards but is kind of a joke right now.
It looks very slow. Is the GPU even much faster than a CPU only render?
It was over twice the speed of his 32-core CPU.
@@thinkingoutloud3757 yes, but that's with two $1000+ 2080 ti's. Still no match for octane or redshift in terms of speed, but certainly a good start.
Been hearing about Arnold GPU coming for a couple years now.. when??
We don’t make Arnold. We just report on it. We will break news here though. Stay tuned!
Could you be so kind in writing your computer specs please?
Where do you get the money to buy all these super-expensive render programs (you seem to have them all) other plug-ins, and on top of that the full version of C4D AND a beefy machine!? It must be atleast $20,000
Well if you are making 65k+ as an animator/modeller, 20k investment for a few years of work is relatively manageable.
@@jcdentonunatco I doubt the lob pays that much...
Is amd card support this
Sir sir how to enable gpu or cuda with pro render in cinema 4d
it enabled by default...
When I try to render it say there is no cuda enable wich support pro or octen render
@@Fahadkhan-sh9kg What graphics card do you use?
say your hardware at the beginning of the video please
You can see our hardware here... greyscalegorilla.com/resources/
anyone explain me what is public beta means ?
It means that Arnold GPU is still in beta but can be tested by anyone 👍
this can work with a gpu AMD?
No.
wow wow wow wow wow
octane is octane :D
It’s the same for nvenc it’s never looked as good as cpu so it never took off
GPU time to first pixel is NEVER going to be as fast as it is on the CPU. CPU rendering in Arnold already blows away most GPU renders in terms of interactivity.
Do you know the technical reason why out of interest? Naively I would have thought the GPU would be faster as in games they render hundreds of frames per second with virtually no lag
@@BENBOBBY My two cents: Firstly you'll have to load everything from disk into ram and then again have specialized a structure to pack data and send to the GPU which takes up some more time. You would have to do this everytime you change something in the scene. Games are faster because they use a different technique called rasterization. With ray tracing or path tracing there is an issue of thread divergence where more scattered the rays are the slower the performance. GPU threads run in lock step and if there is scattered data access or different parts of the code executing for different rays then you would see really bad performance
@@25knightcrawler Sounds logical, thanks! Seems like a combined CPU and GPU renderer would be ideal then to have the best of both worlds.
Sorry i mean Laughing Head Model :)
most renderers now a days tend to use one control for their glossy,reflection,AA,motion blur/dof,GI etc. at first I was not happy that I cannot control one aspect of it but the reason for it is that it will became much faster when your test and for final settings just change one setting even corona and vray works like this... this method is the reason why people like corona because you can control your render quality with just the noise value. even Clarisse' path tracing renderer works this way just one parameter to control the noise quality.
Hold on Vray ! 😂
you can't just switch mid render from CPU to GPU rendering in a second you clown because it's a whole different memory stack and calculation method.
waiting 10 seconds is a drag!? boy you know nothing of gpu rendering. 10 seconds time to transfer data to the gpus is very Very fast.
As a long time Octane render user trust me i know. Its one thing dealing with small scenes its another when scenes gets gig's heavy. though the rendering itself may be fast often the updating of info and data can severly bottle neck GPU workflow. because all the gpus have to be updated then re-synced. so again 10 secs is nothing be thankful.. =
You have to be incredibly stupid to say this lmfao
Rip octane
rip
Rip v-ray too (because it's now 4 times cheaper to use Arnold than v-ray for me). A few years ago it was the exact opposite.
Not impressed, my 5820k is faster then 2 980 ti`s. Takes forever to get a clean looking image when using Arnold gpu, guess is just for lookdev for now.
Sampling needs a TON of work. Hopefully they'll solve it soon.
@@cgpov they cant Chad unless they write their own GPU implementation from the ground up. Something a lot of people are not talking about (we discussed this internally during beta) was the issue with Optix API for ray-tracing does not allow for local sampling. It cant do more samples on GI/Spec/etc. This is why it's forced to run as a pure path-tracer randomly firing only primary rays. They either need to find some crazy workaround or Nvidia needs to further develop the Optix API to allow for local sampling. Both are a very tall order.
@@TheArtofSaul Thanks! Yah it is a bit disappointing for sure. I'd be stoked if Redshift released 3.0 and it magically fixed the issues I have. I'd also love to see RS improve its UX overall. There are so many small annoying things that add up in RS.
Chad Ashley Big things got announced yesterday at GTC, random walk sss, multiscatter volumes, automatic sampling, much higher trace depths, shader sdk, and much more. The big one is the new real-time Redshift RT mode, basically like unreal engine or blenders eevee on steroids for viewports and even renders. Hopefully Nvidia puts up the talk soon. 3.0 initial builds are supposed to drop next month. (April) , just put in those wish list requests in the forums and the devs are very adamant about fixing things and adding features.
The gpu rendering still suck till today ... uses to much memory for such a simple scene with 4k texture ... i switch to octane and its not using nearly half of my 6gb gpu memory ...
If you are using R19 and want to avoid problems and waste of time just stay away from arnold, the last version I have used was so buggy, the 2.5.0 seems better but still I have a lot of problems of crashes, preview of materials and the GPU with ipr doesn't show anything after 1 min of calculations for a 1 million polygons scene , not very impressive...
With R20 however, it's better and GPU seems to work correctly but still I get crashes too so I won't recommand arnold as a main renderer, vray, redshift, thea and octane are far better
I love Arnold and have been using it for years. I dont have a big GPU . I actually need a CPU renderer. I guess what you are saying is, if you want to use a GPU renderer, you wont recommend Arnold. Or are you saying stay away from Arnold regardless because I wouldnt agree with that. its been solid for me for years and still is. (Cpu)
If you need CPU renderer I beg you try Corona, it's 45 days free full function beta, it's monstrous! @@D0nKeY
These scenes are so light not really production stuff mental ray would be just as interactive 2years ago with that Geometry and deferred render setup, buy Xeons,
Оооооооооочень мееееедленннннныыыыыыйййййй реееендееер
somewhere between 8 and 16 huh? Then I am out of your league with measly 4 cores.
Did you mention what gpu is in your system?
Changing the number of samples and activating or deactivating adaptive sampling between cpu and gpu isn't very usefull to convey how both methods compare.
You didn't show the difference between previous randomwalk SSS and this new version.
Sorry but this was a pretty bad first impressions video, I still don't have a clear idea about this gpu beta aside from the long time to load the scene. 👎
2 x 2080ti's thanks for watching!
EEVEE FOR FREE!!!