Hello guys,
in my quest to reduce render time of a specific scene, I've greatly improved thanks to Lee Griggs my cpu render by reducing AA sampling and numbers of lights. Now looking forward, gpu rendering have a reputation to be an even faster solution, at least that what they say but it is quite the oposite here in my case. I've made a few tests at different samples, lower samples with higher adaptive sampling, update my Nvidia drivers, pre-populate gpu cache again, my render time for a similar grain level is still greater than the cpu render time.
machine: 8-cores i9-9900k, 128Go Ram, Nvidia RTX 2080 Ti
software: cinema 4D R21, c4dtoa v6.0.3.1
Arnold lights: 1 sky samples at 3, 1 quad samples at 1,
other lighning: 1 object (big cylinger) with hdr image
cpu render settings = 10:29 min render time
Camera AA = 5
Diffuse = 3
Specular = 3
gpu render settings = 15:55 min render time
Camera AA = 30
same time if I Enable adaptive sampling and put AA samples max to 30
@ 30 samples, gpu renders are slightly grainier than cpu's (with settings above). Anything below 30 is definitely out of comparison. I've noticed that for the same time, Camera AA of 5 with max sampling of 30 is grainier than only camera AA set to 30.
1- What am I missing as per optimisation of the scene or render settings in order to shave render time?
2- GPU render seems to be a lot darker than CPU's is it normal?
I've attached an A/B comparison screenshot with (A) cpu render and (B) gpu
If you are going to use adaptive sampling, you need to reduce the Camera (AA) samples to 4 or 5. Otherwise, your total samples are crazy high. Try adjusting the Adaptive threshold setting lower (.005 for example). If you are still getting grain, then increase the AA samples max to something like 60.
After reading through one more time, your "other lighning: 1 object (big cylinger) with hdr image" could be contributing to the problem. Are you using Emission?
Yes, I do use a 0.7 emission.
This was my way to reduce the number of lights and save render time on cpu setup.
So you think I would be able to lower the samples if I did not have have an object with a texture emission?
Should I place the hdr image directly in the skydome without emission instead?
P.S. Just to specify, my main interest here is to save render time, the amount of grain I get with 5 AA and max 30 is satisfying.
Alex is right, Camera (AA): 30 is crazy high with or without adaptive sampling.
With such high Camera (AA) you probably wouldn't need specular samples: 3. Can they be 2?
I recommend you read through these tutorials related to removing noise.
https://docs.arnoldrenderer.com/display/A5AFCUG/Removing+Noise
Do you have a screenshot of the lights in your scene? Do you need so many?
Hello Lee, I was under the impression that specular and diffuse was not influencing GPU renders since they are grayed out.
Just performed a GPU render without this object containing an hdr image composed with 0.7 emission. Though I've added it to the skydome but without emission.
I've recorded the exact same render time.
CPU vs GPU - any ideas for enchance? Now my CPU (i7 6700k) is better than RTX 2070 Super more than 2 minutes per frame with the same noise scene level.
Can you ask it in another question with more information about the scene with images: settings used, timings, log files, etc so that we can better help you?