Hello.
It was a lot of fun to play around with this.
I’m running 2 GTX 970s using SLI, so I can render 2 tiles at the same time.
The coolest thing to do is to set the tile size to 480x540, because then both GPUs work at the same time and each one is rendering exactly half the image.
But what surprised me is that this is NOT the most efficient way to render!
By lowering the tile size I could cut off a few more seconds.
480x540 - 50 seconds (divides image into 2 even pieces)
1024x1024 - 1 min 27 sec (only uses one GPU)
512x512 - 52 sec
256x256 - 49 sec
240x270 - 47 sec (divides image into 8 even pieces)
480x270 - 45 sec (divides image into 4 even pieces)
240x540 - 46 sec (divides image into 4 even pieces)
128x128 - 1 min 1 sec (now it gets slower again)
My optimal tile size is 480x270 at 45 seconds!
That splits the image up into 4 pieces, allows both my graphics cards to work at the same time, and lets each card render 2 pieces.
But there was a very noticeable difference of 1, sometimes 2 seconds, depending on if these pieces were horizontal or vertical.
This was really interesting to me!
Does someone know why this is the most efficient way to render?
Thanks for reading!