1
$\begingroup$

recently I did a check/test in cycles, the result two images with exact same scene, and exact same format size (1247x848). The first image which is a viewport render took 1 minutes and 45 seconds to render, the other one which is the F12 render took 5 minutes and 39 seconds to render, and I think both renders look pretty much the same quality, and sometimes viewport rendering has slightly better quality! I really don’t understand why. In settings I have 55 samples in viewport, 200 samples in render (but somehow final render with 200 samples doesn’t look better, it only takes more time to render). I like how renders look in viewport render, I could save my renders as screenshots, but the reason why I don’t do that is because the format resolution would be smaller, and because regularly I post-process my renders in DaVinci Resolve program to adjust exposure, gamma and color, most of the time my final renders are saved in OpenEXR format.

I'm adding both renders images down below.

By the way I’m using an old PC, rendering with processor Intel i7. I just want to learn more. what I’m ignoring/missing?
Thanks.

5min39secs 1min45secs

$\endgroup$

1 Answer 1

2
$\begingroup$

You didn't say which version of Blender you're using and it matters, but I'll cover both cases. In either case the main issue is that adding cycles Samples only improves render quality to a point. After a certain number of cycles, your render doesn't get any better.

  • Blender before 3.0 uses Cycles, the original implementation of the render engine. If you're using Cycles, then it will iterate as many times as you have samples. So, in your case, since you have the number of render samples set to roughly 4 times the viewport sample, your render will take 4 times as long, approximately. (It's not a directly linear thing because other factors come into play, but it's close.)

  • Blender starting in 3.0 uses Cycles X, a newer implementation of the render engine. If you don't set a Noise Threshold, then Cycles X will similar behavior to cycles, in that it will compute as many samples as you have specified, and, again, it will take roughly 4 times as long to compute your render as it would take to compute the viewport.

There are a lot of differences between Cycles and Cycles X, but the one that matters to this discussion is Noise Threshold, a concept that doesn't exist in Cycles. It matters because Cycles X will stop sampling if it thinks that the noise has gotten below that threshold, but only if you Noise Threshold set.

So with Cycles X, you can set the Noise Threshold to a relatively high value for a render that won't have much noise and Cycles X will stop when it reaches the point where no more improvement (at least in noise quality) will come. You can do that for both the Viewport Display and the Render calculation; but the defaults are for a much higher threshold in the Viewport, and the threshold stops cycles when the noise goes below the threshold; so again, all else being equal, a render might take much longer but not produce a much better result simply because of how the Noise Threshold is set.

$\endgroup$
4
  • $\begingroup$ Thanks, that solved it. I set noise threshold from .01 to 1, and now it took 55 seconds to render, with f12. I'm using blender 3.1. Thanks! Peace. $\endgroup$
    – Emet Derek
    May 26 at 3:57
  • $\begingroup$ @EmetDerek you're welcome. Happy blending. You might want to read What should I do when someone answers my question? and perhaps accept my answer $\endgroup$ May 26 at 13:41
  • $\begingroup$ I apologize, I thought accepting an answer was already involved when I click the arrow-up... Answer accepted, done. Thanks again. Peace. $\endgroup$
    – Emet Derek
    May 27 at 1:01
  • $\begingroup$ No need to apologize. Stack Exchange is different than other forums and takes getting used to. Enjoy your weekend. $\endgroup$ May 27 at 4:30

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.