I'm asking this theoretical question because obviously I haven't tried it. Is 4 GPU's of 4 GB better than 2 GPU's of 8GB specifically for Blender rendering in Cycles since the app allocates a tile to each GPU in the process? Did anyone experimented or knows which combination is faster given that the motheboard, CPU and memory can handle both of them? TY
1 Answer
"Better" is not the question... it is the wrong word to use, as it depends.
4 GPUs will render 4 tiles simultaneously so rendering an image will be faster than with 2.
Memory: GPUs cannot share vRAM, once you hit the limit the GPU will fail.
So it all depends on the complexity of the scene you are trying to render, or how long can you wait for the render to be done.
-
$\begingroup$ Let's say I have a scene with 10mil vertices in it. Can four 4GB gpu's render faster than two 8GB gpu's? $\endgroup$– DanielJan 9, 2020 at 13:35
-
1$\begingroup$ It is not the number of vertices only, there are multiple factors that determine the resources needed to render an image: modifiers, volumetrics, particles, simulations, size of images used as textures, complexity of materials, number of samples, bounces, render size, denoising, to name a few. $\endgroup$– user1853Jan 9, 2020 at 13:38
-
1$\begingroup$ Test, test, test... $\endgroup$– user1853Jan 9, 2020 at 13:48
-
1$\begingroup$ Depends on a lot of factors including which GPU models you are considering. If you compare 4 low end graphics cards the may end up being slower then only two high end. Scene complexity has a huge role too, some may benefit more from 4 independent cards, other may require more memory, so this question is not really answerable as it stands. $\endgroup$ Jan 9, 2020 at 13:54
-
1$\begingroup$ Speculative questions and hardware recommendations are hard to fit within the format of this site as they generate discussions and opinions. A forum format or social media work better for that. Try blenderartists.org $\endgroup$– user1853Jan 9, 2020 at 14:02