Reading this, I am not sure if everyone is talking about the same sort of tiling...
It sounds to me as if the luxcore-tiling uses dynamic sampling conditions for each tile?
As far as I am aware, cycles just splits the rendering into segments, but otherwise renders the same. The (quite significant) performance differences found are being explained by different overhead, memory and cache usages. Read e.g. here, here, here or here. There are also reports of the GPU-behaviour being different with 2.8 (here), maybe in conjuction with the new CPU+GPU feature and the opposite behaviour of the two regarding tile size?
Tile by tile Rendering
Forum rules
Please upload a testscene that allows developers to reproduce the problem, and attach some images.
Please upload a testscene that allows developers to reproduce the problem, and attach some images.
Re: Tile by tile Rendering
Yes but it is optional, when disabled, it will work like Cycles (i.e. a constant amount of samples for each pixel and tile).
Re: Tile by tile Rendering
This means to me, that adaptive sampling could be improved a lot. By the way my experiences also show this.B.Y.O.B. wrote: ↑Sun Jan 06, 2019 7:13 pm For example if you have an object which is hard to render, but doesn't cover much space in the image, surrounded by background (easy to render).
In this case full image sampling would waste a lot of samples on the background, even with adaptive sampling, while tiled rendering would stop rendering the background after the second pass.