I have a new gpu and i was testing some options in blender with cycles and i saw that rendering a big single tile is faster, not just againts progressive refine, but faster than all other tile sizes when using optix rendering. In this picture, the render on the left is a big tile and the one on the right is progressive refine.
This seems like a bug to me. I don’t have an RTX/Optix card, so I can’t test that, but maybe when i get home I can try it with CUDA and see what I get.
Have you done sample/second measurements? For example, how long does it take to render 100 samples either way? It could be very helpful to measure that sort of thing directly, and with various sampling/light path/ render performance settings. Sometimes, a single checkbox can make a big difference in render performance. This goes a long way to help identify the cause of the slow-down.
With the big tile it doesnt update in real time but it updates every 5 secs. And it always jumps from 0 to 32 and then to 64 and on. But i have compared both, and depends on the scene but in this case you can see the ammount of samples, and the progressive refine has 13 samples in 36 sec when the big tile has 32 in 31 secs. Btw this was rendering on a 2080ti if you wonder.
How is progressive render setup in cycles vs tiling or one big tile? I’m working with cycles, not blender.
Please don’t dig up 3 year old topics. The way Cycles renders changed in Blender 3.0, see Reference/Release Notes/3.0/Cycles - Blender Developer Wiki.