I have a very simple scene consisting of a wireframe cylinder inside a reflective cylinder, the camera travels down the tunnel with a light. It is set up with a cryptomatte pass through and is 4k resolution using eevee with screen space reflections enabled running over 64 frames. I’m running a ryzen 7950x CPU and NVidia RTX 3090 GPU - massively overkill for this scene.
I followed this post regarding multiple instances/GPUs EEVEE MultiGPU? Yes, you can and made a console application to set up multiple processes and timed the execution through code. I also ran the batch script provided in the link and got similar times to verify my code wasn’t doing anything weird. Here are the results of the render time against instance count:
| Instance count | Render Time (seconds) |
|---|---|
| 1 | 200.77 |
| 2 | 147.69 |
| 3 | 103.99 |
| 4 | 93.89 |
| 5 | 91.9 |
| 6 | 90.57 |
| 7 | 89.92 |
| 8 | 88.95 |
| 9 | 89 |
| 10 | 89.01 |
| 12 | 89.9 |
| 16 | 90.64 |
Regarding resource usage memory was fine in both system and GPU and GPU utilisation was maximum 70% very briefly. CPU was spikey and started maxing out past 8 instances which is where you start to see the render getting longer again which makes sense. I also have another GPU which when combining both GPUs didn’t bring any performance gains.
I do have limited knowledge of the render process so this is where I have questions. The render time decreases very fast with increasing instance counts to begin with then tails off with very little gain from 5 to 8 instances. Nothing is maxed out on my machine so what could be the cause of this?
For this use case would the only way to get better results be to get a CPU with more cores?