What are "samples"?

[Cpt. F****ace]

Hello there.

I am starting with blender and I still don't understand "samples". When I try to google "what are samples blender" or just "samples" there is no defition. For example I made simple animation and tried 50samples and 200samples. No difference other than render time.

Best Regards.

[still__alive]

From this article on Blender guru: https://www.blenderguru.com/articles/4-easy-ways-to-speed-up-cycles/

4. Reduce Your Samples

Put this one in the “duh” category, since most people already know it. But samples are the biggest time sucker of all, so I need to mention it.

What are Samples? Samples are the noise that appear as your scene is rendering. In the render panel you define the number of samples, and then blender stops once it reaches it. The more samples, the clearer, but longer your render is.

More samples are generally good. But there comes a point where more samples does almost nothing.

It might depend on how complex your lighting setup is and stuff, but with low samples you can see noise, sort of like the top half of this image: https://s3.amazonaws.com/blenderguru.com/uploads/2014/06/promo-673x420.jpg

[Cpt. F****ace]

Thanks, but I already know it... but I am too dumb or just making things too hard. Samples are noise? Or less samples cause this noise?

Trying to ask it from other view: what does rendering engine do with 50 samples and with 500samples? Like trying to fix lightning on each pixel 50 or 500 times?

Anyway thank you for response!

Best regards.

[still__alive]

Originally posted by Cpt. ♥♥♥♥♥♥♥♥:

Thanks, but I already know it... but I am too dumb or just making things too hard. Samples are noise? Or less samples cause this noise?

Trying to ask it from other view: what does rendering engine do with 50 samples and with 500samples? Like trying to fix lightning on each pixel 50 or 500 times?

Anyway thank you for response!

Best regards.

Well I don't understand all the ins and outs of it, but I think that more samples does equal more time spent on calculating what each pixel is supposed to look like. For some renders maybe there is not enough material and/or lighting information to make more samples worth it. For some renders you might need more, for some you might need less.

[tripmix]

https://docs.blender.org/manual/en/dev/render/cycles/optimizations/reducing_noise.html?highlight=samples

There is also this firefly remover here:

http://www.blendswap.com/blends/view/85470

Once downloaded and extracted you can then append the node setup to the compositor and it will reduce noise some noise.

You can also adjust the clamp direct and indirect settings but if you have to have them too high to produce an acceptable result then there is other issues with your scene.

Are you using mesh lights?

[Mr Chappy]

Hi,

AFAIK this is roughly how samples work....This is the best explanation I can give, I'm not an expert and apologise in advance for anything I may have got wrong, but this should hopefully be roughly right...

In real life light rays bounce off surfaces and some of these rays hit your eyes which is what you see.
In Blender it is sort of the opposite way round during rendering. The light paths are traced(projected) from the camera(the scene's eye) outwards. I'm not sure but I think it may be one ray per pixel of camera resolution, may well be wrong about this bit so don't take this as gospel but it does make sense if it is!
Note that the camera is not lighting up anything, your lights do that! The light path is however traced backward from it's end point(the camera) back towards it's origin. This is a sensible way of doing things if you think about it as it only processes the light paths vital for creating the image. Imagine the PC power you would need if you tried to calculate everything and not just what was needed!
When a ray makes contact with a surface it gets deflected. This ray can be deflected in more than one direction by the surface, the type of surface it hits will affect the spread of different possible directions it is likely to be deflected in.
Each sample will follow the ray along one possible path of deflections for the set number of bounces(deflections) this ray will make before it expires(defined in your render settings under “Light Paths”).
So if you have one sample you have one possible outcome from the potentially infinite number of possible outcomes.
Each sample you add will follow the ray down a different possible path of deflections and therefore possible outcomes. So having 5000 samples means that you get data for 5000 possible paths taken by that particular ray over X number of deflections(bounces). These results are then combined into an average that give a much more accurate result than if only one or one hundred paths had been explored.

You will see an option for "branching" in the render settings. This means that it sends out a ray, then upon a deflection the ray splits(branches) into X amount of rays, each of which then split into X amount more rays upon the next deflection, meaning you now have XxX number of rays and so on exponentially. This is a different method of achieving the same sort of thing, I can't say what it's advantages over the other method are though?

So AFAIK fireflies are basically caused by a lack of quality data, resulting in a poor outcome for that pixel that is at odds with those around it. Certain things like caustics, highly reflective surfaces, and low resolution or low sampled Hdri's, are more likely to result in a poor data set than others.
These therefore need a higher sample rate or clamping(as discussed above) to either improve the data or compensate for the lack of it.

That is my best shot an explaining this with my current knowledge, hope it helps a bit! :D

[*P0P$*FR3$H3NM3Y3R*]

To expand on that: you can actually render an image at lets say 100 samples and get a result. If u render the same image with a different seed at another 100 samples u will get a somewhat similar (not identical) result.
Mixing these images 50-50 you wil get an image that has the precision of a 200 sample image. They actually stack. You could do 10 different seeds at 100 samples mix the 10 images at 10% each and u get a 1000 sample image ;D

I dont know if that helps..i just find that interesting that it works that way.

So generally the more samples the less noisy the image will become. However with some tricks u can get a nice clean image without having to go overbaord with sampling
Sometimes its advisable to rather increase bounces instead of increasing samples.
Especially for animations hitting the sweet spot between image quality and render time can be tricky.
Doing test renders at lower resolution can help compare different render settings in terms of their required time. (50% image size = 25% time needed per frame in full res)