Hey there Blender Community,
I have a python script set up that allows me to iteratively render animations with minor adjustments each time I render. Effectively, the script sets up a scene, then renders the scene for a set number of frames before repeating the process again.
I wanted to try and accelerate this process, and have at my disposal a GeForce GTX 1080. I enabled GPU compute, set the Cycles Compute Device option in User Settings to CUDA (ticking the box next to the GTX), and was surprised to see that there wasn't really an increase in rendering speed--that there was, in fact, a decrease.
For 5 animations of 72 frames each on the CPU, the total rendering time was 7 minutes and 12 seconds (an average of ~1 minute 40 seconds per animation).
For 5 animations of 72 frames each on the GPU, the total rendering time was 9 minutes and 17 seconds (an average of ~1 minute 51 seconds per animation).
I reckon I'm probably doing something wrong. I am using Cycles in Blender 2.79, and would be much obliged for any advice. Thanks in advance!
UPDATE I: I have benchmarked the GTX using Mike Pan's BMW Benchmark Scene; the render clocks in at ~ 1 minute 4 seconds. In terms of my own animations, I have discovered that adjusting tile size decreases the rendering time per frame from around 1.51 seconds to approximately 1.15 seconds. I'm still hoping that there are ways to decrease this even further.
UPDATE II: The scene is a very simple scene, with only a few rigid body objects, all diffuse materials and one light source. Even on a slower CPU, the average rendering time per frame is still less than 3 seconds. I am hoping to decrease the rendering time per frame to half a second (.5 seconds) or less.