Slow down by denoiser?

[Micha]

Hi,

I wonder why the denoiser cost so much rendertime. For example I render a product shot (6400x3600):

1m:15s no denoiser (0,03)
1m:49s with denoiser (0,03)

I found that it can be useful to render without denoiser, but use a finer noise threshold.

1m:50s no denoiser and 0,02

I did an other test and it was the same - denoising 0,03 is nearly so fast like 0,02 without denoiser.

It could be nice if the denoiser wouldn't increase the rendertime.

[piotrus3333]

bucket or progressive?
denoiser - vray or nvidia?

[Micha]

I did a new test at 1920x1080. VfR5. product shot for animation

Cuda, Progressive, noise 0,015
21s no denoiser
25s V-Ray denoiser

Cuda, Buckets, noise 0,015
22s no denoiser
27s V-Ray denoiser

RTX, Progressive, noise 0,015
30s no denoiser
32s V-Ray denoiser

RTX, Buckets, noise 0,015
29s no denoiser
36s V-Ray denoiser

We see, the V-Ray denoiser cost some extra time and RTX is slower than Cuda. Bucket is a little bit slower than Progressive, because the large buckets and not all buckets are used until the end.

I hope this test show how to get speed and I hope the code can be optimized so that bucket mode, RTX and denoiser doesn't cause a slow down. This nice features needs to be used careful at the moment.

[piotrus3333]

vray denoiser for animations is designed to be calculated after you have the whole sequence rendered.

if you are denoising every frame separately than push the denoising to after frame is done in progressive mode (effects update rate option or something like that)

are you using lightcache for secondary gi bounces? depending on hardware you might be faster with brute force if frame time is so low.

[piotrus3333]

and if you want to denoise every frame separately for whatever reason - give nvidia denoiser a try. it did work ok for me.