I am planning to create a very complex scene (vase of flowers) and I am interested to know - before I invest too much time - whether it will be possible to render it.
I have a fairly powerful hardware setup (including a top-end GPU) and a reasonable amount of memory. Even so, I found that one of my early experiments ground to a halt on the GPU and using an automatically-determined tile size. Re-rendering on the CPU or interestingly, opting for progressive refinement, was much better.
I have two questions:
Is there an account anywhere that explains how rendering tasks are divided between CPU and GPU ? For example, I would like to understand the implications of using different tile sizes.
Is it true to say that virtually any scene can be rendered if broken down sufficiently (either into layers or separate tiles) or are there some scenes that are simply too complex ?
Can you provide the exact details of your hardware, especially the GPU (and how much memory it has)
Generally tile size of 256x256 or 512x512 provide good performance with main build of Blender 2.79b. In later versions they made tilesize options almost not important.
Also I recommend if you can find a photo of what you’d like to achieve, it would proivde us guidance for the response.
Initiallly I have to say that you will be able to render such scene.
Additional question, do you want to animate the scene or just single render at 4K (or higher res)?
Are you testing the limits of what is renderable with your hardware ?
Because I’d say that with clever adjustment optimization and trickery, you can manage to render a lot’s of stuff.
It’s not because something looks complex and as a lot of detail that it should be very highpoly and hard to render.
The image above took 25min to render 100 samples (peak memory: 807 MB)
It is structured as a particle system instantiating a group of mesh object (petals).
The object above is intended to be a component of a larger scene. Eventually I would like to create something approaching the complexity of the example below (which is a painting).
However, I don’t understand the implementation details of Blender well enough to know how best to structure my scene and would be grateful for any advice.
Should I use layers, for example ?
Are there other ways of dividing the task so as to process a large amount of data ?
Please see my earlier response to Grzesiek for further details of what I am trying to achieve.
I would be grateful for any tips on the ‘trickery’ you talk about.
Hey, that’s more clear with a picture, and it’s indeed a complex scene you’re planning to do
Very nice render BTW ! It’s starting quite good !
Anyway, your render times and memory are quite high …
Are you using GPU ? What are your render settings ( bounce , title ? )
About optimization and strategies to handle such a scene, indeed you can’t do much stuff like , faking some part of the scene with image instead of mesh, but :
You may start by blocking out the scene, create a rough composition with simple, low poly models. And set camera angle + some rough animation shots if you plan to do animation.
That will tell you how much detail will be needed , and you may, at that moment try to reduce geometry. Or use some normal maps or bump maps to fake high poly count. These petals are full geometry ? they seem to have some kind of solidify , that may not be needed depending on the distance of the camera.
If you’re filling 807Mb with only that ( if there isn’t texture involved) , then there is an issue.
Also, can you do a clay render with default principled shader ? That will tell how much of shading is causing this long rendertime .
Good luck ! I’m looking forward your progress on your scene !
That’s such good feedback. Thanks.
I’m going to look at the things you suggest but it may take me a little time.
I really appreciate your suggestions about a ‘blocking out’ approach to the scene and also about doing a clay render. All makes a lot of sense.
I’m glad you like the render. You’re right - it is full geometry with solidity etc.
It would be great to see if I can distil it into some sort of texture or pre-baked component.
Here’s the latest render by the way. This is all about forms, irregularity etc and less about the shader. This one isn’t solid by the way and no subdivision etc.
Hey awesome !
It’s true that I’d like to block things early, but it’s not always necessary. You can still experiment an try to optimize when you’ll put all your tests together to form the final picture.
And if you’re working on a still image it’s more about saving memory rather than time ! A little night of computation is worth a nice image
I just noticed that you render image in 3077x2500, it explains a bit why 25mn, even if I think it’s still possible to render faster. Do you use cycles defaults settings ?
Feel free to poke me once in a while, it will be great to see the evolution !
Keep up the good work !
I agree. I am very happy to put in the rendering time - one (two) nights. Not a problem.
I just don’t want to get to the point where I simply can’t load the model into the machine.
In the meantime I will do some research on limiting polygons (decimation? etc.)
To be honest, I hadn’t realised my image was that large. It’s a setting inherited from a different final render.
Thanks for your help. No need to respond to this message. I’ll post some more when I have something new to show.
