Sample Overrides vs Automatic Sampling?
prompt-height last edited by Dr. Sassi
I am trying to figure out how to best configure Redshifts Samples for faster speed and was wondering how often other people actually go inside of the Sample Overrides instead of just using Automatic Sampling and turning the threshhold slider up or down?
Is it usually worth it to go inside of the overrides and turn everything off/down that is not being used and cranking everything way up that is being used or how do I go about this? I also just recently found out about the AO views inside of the RS render view, so do you check there what is actually being used/how noisy the pass is and then going inside of the overrides to find out the settings there?
Dr. Sassi last edited by Dr. Sassi
Thanks for the question.
If you feel comfortable doing it, take advantage of it. I think it is worth the effort.
I worked on a small clip this week, and a single-frame render resulted in 30+ minutes. After some exploration, I brought it down to 2.5 (average) minutes per frame.
At the same time, no real visible degradation of the result can be seen without pixel-peeping. (Pixelpeeping, if one has to stop earlier, based on longer render times, the "story" of a clip might not be where it could be, and the attention goes to the problems instead of the content.
I do not work for pixel-peeper, which has saved me a lot of grief. The key is to find what is visible while watching and how much one can catch the audience's attention away from tiny things like noise.
When the brushstroke becomes more interesting than the picture, we have a different audience.
That is not easily generalizable; otherwise, I would have filled an hour show with it and be done. Every training I watched about Redshift 3D had this theme, more in earlier years, to state a subjective observation.
However, from my point of view, the optimization starts much earlier. I would say in carefully creating materials, textures, models, and pretty much anything. Fix it later is not working.
A year ago, I had one case on the desk that seemed artificially slow, and setting up the material made it many times faster, which started me to look closer.
In other words, yes, one can speed up things in many cases, but initially, it might take longer until one establishes an eye for the different qualities that indicate a time trap. Or one creates ideas for optimizing things before anything is put into place.
How much time investment and saved time balance each other or go even in one or the other direction is hard to predict. I can confidently tell that gained experience will pay off in the long run.
That experience might help to have less hardware investment to get the same result. How much your time is valued, or your render farm costs you is an individual calculation. Nothing anyone can tell you, but it boils down to that.
I can only encourage you to take idle GPUs and run tests. Even if it increases the CO2 footprint now, it helps to reduce it in the future.