I created super complicated scene using 8gb out of 6gb of vram. On the way of trying to reduce memory usage I saw many times my gpu rendering only one tile, with all samples, and skipping rest of the tiles with message at the end "out of memory". Right now it hit me that I can set a tile to a size of full render. It would probably render quicker with 256 by 256 tiles. It just finished... and looks normal. Has anyone done it before? Or is it a common technique only I didn't know? What are the cons? Does it differ from properly rendered tiles?