I performed multiple tests with both GPU and CPU and here are the results:
| Width | Height | Scale | Device | Tile Width | Tile Height | Time 1 | Time 2 | Time 3 | Time Avg | Notes |
|---|---|---|---|---|---|---|---|---|---|---|
| 1920 | 1080 | 50% | GPU | 64 | 64 | 01:57.42 | 01:57.28 | 01:57.46 | 01:57.39 | |
| 1920 | 1080 | 50% | GPU | 128 | 128 | 01:39.47 | 01:39.47 | 01:39.69 | 01:39.54 | |
| 1920 | 1080 | 50% | GPU | 256 | 256 | 01:36.60 | 01:36.59 | 01:36.65 | 01:36.61 | |
| 1920 | 1080 | 50% | GPU | 512 | 512 | 01:37.84 | 01:37.61 | 01:37.56 | 01:37.67 | |
| 1920 | 1080 | 50% | GPU | 120 | 90 | 01:40.40 | 01:39.96 | 01:40.54 | 01:40.30 | 8:6 of scene |
| 1920 | 1080 | 50% | GPU | 240 | 270 | 01:34.03 | 01:34.05 | 01:33.96 | 01:34.01 | 4:4 of scene |
| 1920 | 1080 | 50% | GPU | 480 | 540 | 01:36.19 | 01:35.99 | 01:36.65 | 01:36.28 | 2:2 of scene |
| 1920 | 1080 | 50% | GPU | 960 | 540 | 01:38.86 | 01:38.75 | 01:38.75 | 01:38.79 | Full scene |
| 1920 | 1080 | 50% | CPU | 8 | 8 | 03:15.58 | 03:15.82 | 03:16.11 | 03:15.84 | |
| 1920 | 1080 | 50% | CPU | 16 | 16 | 03:16.24 | 03:17.23 | 03:16.15 | 03:16.54 | |
| 1920 | 1080 | 50% | CPU | 32 | 32 | 03:18.68 | 03:18.26 | 03:18.85 | 03:18.60 | |
| 1920 | 1080 | 50% | CPU | 64 | 64 | 03:24.65 | 03:26.01 | 03:25.22 | 03:25.29 | |
| 1920 | 1080 | 50% | CPU | 8 | 9 | 03:15.49 | 03:16.14 | 03:16.43 | 03:16.02 | 120:60 of scene |
The things I learned are:
When using the GPU:
- Find the optimal tile size for YOUR GPU.
- Divide the scene evenly or process power will be wasted on uneven (smaller) tiles.
When using the CPU:
- Use the “Hilbert Spiral”. “Center” is the slowest.
- Use the fastest power of 2 tile size as evenly diving the scene had little effect since the smaller the tile, the faster it’ll render.
My system specs are:
- Intel Core i7 4790
- AMD Radeon RX 470 4G
- 16GB DDR3 1600Mhz
Hope this is helpful!
Bonus render using Filmic blender:
