justaviking
|
justaviking
Posted 6 Years Ago
|
Group: Forum Members
Last Active: 2 Weeks Ago
Posts: 8.2K,
Visits: 26.5K
|
BOTTOM LINE ON TOP 1: The Denoise performance impact is minimal. BOTTOM LINE ON TOP 2: The visual benefit of turning Denoise on early is zero. BOTTOM LINE ON TOP 3: CONCLUSION: Even though there is no benefit to turning Denoise on early, there is no reason not to
I did two evaluations: a) What is the performance (render time) impact of having Denoise turned on? b) What is the visual impact of turning on Denoise early vs. late in the rendering process?
PERFORMANCE TEST:
I did a 3,000-iteration render using a GTX 1080 graphics card (the desktop system in my signature) and did four tests. Image resolution = 1920x1080
Full disclosure: I did not repeat the tests to get average times
1) Denoise OFF - Leave the Denoise check-box unchecked 2805 seconds - This is as low as it can get because it is really, really "off"
2) Denoise "on but not used" by setting it to start at iteration 3,005 (which is after the render stops at iteration 3,000) 2825 seconds - There is some data collecting that occurs so it is ready when/if the denoise becomes active
3) Denoise ON at frame 2,995 2858 seconds - Here I start denoising at almost the last possible moment
4) Denoise ON at frame 5 2857 seconds - Here I run denoise for basically the entire render process
PERFORMANCE ANALYSIS:
a) All my runs are within 2% of each other b) There is a modest impact on having the Denoise feature "available," but very little impact in actually doing the denoise operation itself. c) Nvidia says it operates in "milliseconds" and based on these results I would say that seems to be true. d) The amount of time saved by denoising "late" as opposed to "early" is insignificant e) If you even potentially risk hitting your "Time" limit before hitting your "Iterations" limit, be sure to set Denoise early enough to avoid the risk of not denoising, assuming you want it on
VISUAL TEST:
For this test, I used 500-iteration renders. a) Denoise OFF b) Denoise on at Iteration 25 c) Denoise on at Iteration 490
VISUAL TEST RESULTS: a) The value of having Denoise ON is very apparent. b) There is no visual difference between turning Denoise on at Iteration 25 vs. 490 c) The Denoise is not a "progressive" operation. It simply works on each frame as a single, stand-alone entity.
I also put the images on different layers in the GIMP photo editor and did a "subtract" operation. It wasn't a mathematical analysis, but I could see zero differences between the two denoised images using that method, which substantiated the visual comparison of the two images.
ONE FINAL COMMENT - WHAT ABOUT THE PREVIEW WINDOW?
There is some overhead in loading up data for the denoiser. While trivial in a longer-running, many-iteration render, it can impact "real time" operations, such as when you rotate or zoom in on a model.
For the PREVIEW window, you should probably have the denoise "Start Iteration" set to something like 10. Maybe as low as 5 if you wish, but not 1. That is an Nvidia recommendation.
The first few iterations are so noisy, even the denoiser cannot save them, and you don't need that overhead (even a fraction of a second) when you are dynamically moving a model and the render keeps starting over at "1" again and again. It would only slow that down without adding any value.
Denoise = OFF Denoise = Start at Iteration 25 Denoise = Start at Iteration 490
iClone 7... Character Creator... Substance Designer/Painter... Blender... Audacity... Desktop (homebuilt) - Windows 10, Ryzen 9 3900x CPU, GTX 1080 GPU (8GB), 32GB RAM, Asus X570 Pro motherboard, 2TB SSD, terabytes of disk space, dual monitors. Laptop - Windows 10, MSI GS63VR STEALTH-252, 16GB RAM, GTX 1060 (6GB), 256GB SSD and 1TB HDD
Edited
6 Years Ago by
justaviking
|