Ramone163 is on to something here...
2 things:
1) Data collection process
2) Denoising processing
What if the '1' -data collection process started later, say after a certain point. You could have the fast 'Main' render, then a later initiated '1'- point of data collection, followed by the ending '2' - Denoising processing, itself.
Like, does the Denoiser, in its entirety, need to be actively collecting data during the whole render?
Because if it could start the data collection process later, you could have the '1'- data collection occur, in say a 1-second's time, followed by the actual 1 or 2-second '2' - Denoising image-affix, all at the very end of the normal-sped render.
This would be very powerful, especially for UHD renders.
Hello World: OctaneRender 4 is here
Forum rules
NOTE: The software in this forum is not %100 reliable, they are development builds and are meant for testing by experienced octane users. If you are a new octane user, we recommend to use the current stable release from the 'Commercial Product News & Releases' forum.
NOTE: The software in this forum is not %100 reliable, they are development builds and are meant for testing by experienced octane users. If you are a new octane user, we recommend to use the current stable release from the 'Commercial Product News & Releases' forum.
You just need to either denoise at end or denoise every 10 to 20 seconds if you don’t have a dedicated GPU to throw at it.
All my tests was with "denoise at end"Goldorak wrote:You just need to either denoise at end or denoise every 10 to 20 seconds if you don’t have a dedicated GPU to throw at it.
980+1080+1080Ti(for denoiser)
We are collecting around 20 render passes data internally for denoising, its lot of data to move around device-host-device, that is why render time is high when the denoiser is enabled. we do have plans for reducing it, by training multiple models with fewer render passes. But first, our focus will be on stability, volumes, memory usage., etc..
@Notiusweb yes we need to collect the data for the whole render time to match the main beauty pass.
Current AI I'm integrating might do better with fewer samples than the XB1 AI, But I need to test it first.
@Notiusweb yes we need to collect the data for the whole render time to match the main beauty pass.
Current AI I'm integrating might do better with fewer samples than the XB1 AI, But I need to test it first.
Okay, just so we don't miss anything, let's walk through this together:Goldorak wrote:You just need to either denoise at end or denoise every 10 to 20 seconds if you don’t have a dedicated GPU to throw at it.
Let's say I have a 1080p render:
150 s/px - NO DENOISE - 3 seconds
150 s/px - DENOISE - 10 seconds (render and denoise data collection) + 2 seconds (Denoise processing at end) = 12 seconds
So my option would be to go to, say 30 s/px denoise, to try and match the 3 seconds regular. But in this case denoise winds up being 5 seconds, but okay, not a big deal, even if animating 1,000 frames, etc.
But then here is the crux....it winds up looking worse than the regular 3 second render for 150 s/px. Not bad, just the regular looks better, fidelity wise.
And mind you, it's only because Octane is so fast that this is happening, I mean, it's not a bad thing that Octane is blazing fast. But, it winds up negating the value of the denoiser, at 1080p +. (At 800 x 600, it is almost instant. )
But so you mention a "dedicated GPU to throw at it". I like this. But how do I dedicate a GPU to denoising?
I tried this and it never made a difference in speed. Is there really a dedicated way to have one GPU strictly manage the denoising such that it makes the denoising faster?
Wouldn't the GPU also be rendering, or is there a way to strictly set it in a denoise mode? Very interested in the procedure if possible, maybe it will help us to have it explained.
Could we opt to have CPU do it, or a separate GPU, etc...Whichever is faster.
THX!
EDIT - @ 720p, denoiser wins almost always. Just > 1080p is when it starts to lose value
Win 10 Pro 64, Xeon E5-2687W v2 (8x 3.40GHz), G.Skill 64 GB DDR3-2400, ASRock X79 Extreme 11
Mobo: 1 Titan RTX, 1 Titan Xp
External: 6 Titan X Pascal, 2 GTX Titan X
Plugs: Enterprise
Mobo: 1 Titan RTX, 1 Titan Xp
External: 6 Titan X Pascal, 2 GTX Titan X
Plugs: Enterprise
- mib2berlin
- Posts: 1194
- Joined: Wed Jan 27, 2010 7:18 pm
- Location: Germany
Hi, are Linux/OSX 4.0 builds in sight?
Cheers, mib
Cheers, mib
Opensuse Leap 42.3/64 i5-3570K 16 GB
GTX 760 4 GB Driver: 430.31
Octane 3.08 Blender Octane
GTX 760 4 GB Driver: 430.31
Octane 3.08 Blender Octane
Would it be possible to use old gpu's as AI coprocessors?Goldorak wrote:You just need to either denoise at end or denoise every 10 to 20 seconds if you don’t have a dedicated GPU to throw at it.
I'm soon going to build a new workstation and have several GTX590's laying around

Win7 Sp2____2xGTX1060 [email protected]____Su17.2.2555 and Su18.0.16975____SketchUp Plugin 4.00.0.17____4.00.RC7 Standalone
It's depends on what GPU is used to denoise. 1080Ti seems to be the best.
Also I have done some tests
150sp and 1080p and 2160p.
Time for my render + denoise seems to be constant from 1080p to 2160p:
https://www.youtube.com/watch?v=yf8zM8p ... e=youtu.be
Also I have done some tests
150sp and 1080p and 2160p.
Time for my render + denoise seems to be constant from 1080p to 2160p:
https://www.youtube.com/watch?v=yf8zM8p ... e=youtu.be
I7-3930K 64Go RAM Win8.1pro , main 3 titans + 780Ti
Xeon 2696V3 64Go RAM Win8.1/win10/win7, 2x 1080Ti + 3x 980Ti + 2x Titan Black
Xeon 2696V3 64Go RAM Win8.1/win10/win7, 2x 1080Ti + 3x 980Ti + 2x Titan Black
Hey itou31, great compare!itou31 wrote:It's depends on what GPU is used to denoise. 1080Ti seems to be the best.
Also I have done some tests
150sp and 1080p and 2160p.
Time for my render + denoise seems to be constant from 1080p to 2160p:
https://www.youtube.com/watch?v=yf8zM8p ... e=youtu.be
So you get, for that BEAUTIFUL render

1) 14 seconds @ 1080p for 150 s/px
2) 34 seconds @ 2160p for 150 s/px
So now, would be interesting to see what would you get with
3) a non-denoise render (Denoising off) @1080p, for 150 s/p,
4) a non-denoise render (Denoising off) @2160p, for 150 s/p,
and how many samples can you get to in
5) a non-denoise render (Denoising off) in 14 seconds @1080p, and then
6) a non-denoise render (Denoising off) in 34 seconds @2160p
Win 10 Pro 64, Xeon E5-2687W v2 (8x 3.40GHz), G.Skill 64 GB DDR3-2400, ASRock X79 Extreme 11
Mobo: 1 Titan RTX, 1 Titan Xp
External: 6 Titan X Pascal, 2 GTX Titan X
Plugs: Enterprise
Mobo: 1 Titan RTX, 1 Titan Xp
External: 6 Titan X Pascal, 2 GTX Titan X
Plugs: Enterprise
I7-3930K 64Go RAM Win8.1pro , main 3 titans + 780Ti
Xeon 2696V3 64Go RAM Win8.1/win10/win7, 2x 1080Ti + 3x 980Ti + 2x Titan Black
Xeon 2696V3 64Go RAM Win8.1/win10/win7, 2x 1080Ti + 3x 980Ti + 2x Titan Black