Because i couldn't find a clear answer on the forum, i ran some benchmarks to have some numbers.
As most of you know :
1/ VRAM is supposed to be really fast, but isn't upgradable, and can run out quite fast.
2/ Windows10 reserves 20% VRAM PER GPU even if it's not the main GPU. Meaning that a 6GB only has 4.8GB usable.
3/ System ram is easy and cheap to upgrade
Octane has an option, defaulted to false, which tells the rendered to use System Ram (Out Of Core), rather than VRAM.
Upon searching the forum, you have people quoting the documentation ("It has a heavy speed impact"), and then some users mentioning ("it's great, and barely noticeable").
But i cannot go on those comments, i need numbers. I build machines for people who use Octane (or other), and i must have precise answers. So i made a small test bench today.
Here are the results
Machine :
i7 8700k
2x 980TI
1x 8gb of DDR4 (PC4-24000) (so yes, running in single channel, so pretty much a worse-case scenario )
Octane Standalone Demo 4.02
Test :
* Downloaded Octane Standalone Demo (https://home.otoy.com/render/octane-render/demo/)
* Downloaded Benchmark Files (https://render.otoy.com/octanebench/ at the bottom)
Opened Scene file, and didn't change any settings.
I choose the "Interior" scene as it's in my opinion the heaviest scene.
Results:
Data:
Code: Select all
Interior DL
VRAM : 13m42
OOC : 13m51
Interior Info
VRAM : 05m00
OOC: 05m03
Interior PT
VRAM : 30m55
OOC: 30m57