So I’ve got this incredibly noisy scene that I’ve been leaving to render overnight over the last couple nights. Now I’ve got two 10,000 sample renders (from different seeds). Would I be right in assuming that if I just layer one over the other with 50% opacity in GIMP, that’d be roughly the same as if I had done 20,000 samples?
Probably just add them together and divide by two (average). Could do this is compositing as well I guess, don’t know about GIMP.
A 50% opacity is fine for a simple blend like that.
If diving by two (average) isn’t easy to do in GIMP:
And if you wanted to average more than 2 images
Set image layers to Add blend mode with 1/2 (50%) opacity (or 1/N where N is the number of images) on top of a black background.
So 4 images would be 4 layers using Add at 25% opacity each = 100% total opacity.
Thanks peeps. I thought I was on the right track, but the results weren’t looking right. Turns out the scene is just really really really noisy even at 20,000 samples. Might have to give it another night or two.
20,000 Samples seems a lot to me.
If it´s not getting better with 2000-3000 samples, something is not right.
There are two worst cases for cycles:
One would be, if your light sources are too small, the other one is that you are lighten your scene only with indirect lighting.
In both cases cycles would be grainy, no matter how much samples you throw at the scene.
Show the scene and lighting setup. Even if you decide to keep on rendering, you could end up learning something about faking and optimizing for the next project. I’ve had projects which would not be renderable if I relied on brute force alone, and had to fake a lot of stuff to get it done in reasonable time.
For indirect lighting where it won’t change, render low quality first and then try to mimic the lighting using direct lights. Time consuming to set up. Depends on how accurate you (more importantly, the client) need to portray the indirect lighting. I’ve had times where I’m uttering “seriously, this can’t be good enough, it’s crap!” 
For indoor room lighting with indirect lights and tons of spotlights, bake out the direct effects and use the result as direct lighting, setting maximum bounces for the original lighting to 1 (or 0, can’t remember). For tons of spotlights I may replace them with big area lights to do the diffuse.
Cycles is still slow when doing many lights, tiny lights, indirect lights, as brute force. We still need to help it out.