I’ve been rendering a simple scene in Blender 3.0; just discovered that rendering time in RTX 3070 and RTX 3090 are same (17 seconds at each at cpu+optix hybrid).
Considering the huge difference in terms of specs & price between two models; confused that how RTX 3090 isn’t much faster than 3070… or do I make mistake in render settings ?
Computer 1: RTX 3070 + Intel I7 11900 + 16GB RAM DDR4
Computer 2: RTX 3090 + AMD 3700X + 32GB RAM DDR4
**Blender project file is attached, may anyone has RTX 3090 help me to render an image in same settings, see if render time differs or not ** project file.blend (1.3 MB)
Don’t use the CPU as an Optix compute device because when used with any reasonably fast GPU it creates a bottleneck. That said, on my 3090 that scene renders in around 2 seconds, or 3 seconds with CPU (3900XT) + GPU, so you may have some other issue with your machine.
So something is a bit wrong. Just for the fun I loaded up the file, the only thing I changed was selecting GPU Compute as Device on the Render Properties, it was set to CPU when I opened the file.
Only using a GPU, I got a render time of 5.68s, thing is my GPU is only a 1070Ti with CUDA set in preferences.
If I set it to OPIX it takes about 1s longer, but then I don’t have a RTX card, so that’s not surprising.
If I select GPU (CUDA) + CPU (5900x) I get a time of 4.96s
However, and here’s the thing, if in the Render Properties, I change the Device setting back to CPU, then it takes 10s.
So double check that setting, my but is you are not even using the GPU to render and it’s a pure CPU render and hence why it’s taking so long.
The overall mistake is the hybrid using CPU + GPU.
CPU is slower, most of the time and because of that, you are slowing down the power of the GPU in hybrid usage, even with Tile Stealing etc…
Fastest way is using optix.
For faster or slower rendering, the tiling in memory is important. 2048 Tile size is sometimes slower as 512 or 256 tile size. I tried a lot with the tile size and it worked the best with 256-512 tile size with 2x RTX 2080 Ti or 2x RTX A6000 (equal 3090) in my case.
thank you all guys for prompt feedbacks.
I’ve cancelled hybrid method, only with GPU get something like 2 seconds rendering time. :=) Besides, after setting hybrid, the renderer is turned to be CPU which caused 17 seconds of rendering time. Fixed the issue, thanks a lot.
