My scene too complex for GPU

Um, I have pretty complex scene, or at least I think so, 'cause Cycles GPU give up rendering it after building BVH. These is what it says in the render port: Men “2045.38M Peak 2045.38M”
It can be rendered in CPU, but you know… another couple of days render time :spin::spin::spin::spin::spin:
I’ve read that you can split up a scene using seperate layers and so render it out, but my scene only contains one character model so that won’t work here.
I also tried packing my textures with no luck.

Is there anything I can do to render this out with my GPU?

Thanks!

You can lower the poly count, or use smaller textures. In the scene tab, at the bottom there’s a area called simplify, check the checkbox and you can specify max subsurf levels and also override children particle numbers, try lowering max subsurf to 1 or 2, and children to .5 or so, see if you can squeeze it into the GPU.

Just curious. How much memory does your graphics card have?

And what size are you trying to render your image to? That’s another thing that can affect the size of the needed memory capacity.

I realised that the problem come down to two things: too many children particles and my enormous texture stacking. I lowered the amount of children particles and it did render!!! Buuuut character model rendered black - probably due to my big texture stacking. Tried to bake the textures but I just get the error message “circular dependency” :frowning:
I’ve never really baked textures before, so I don’t know how to solve this… I can’t find the circular dependency Blender tells me should be there. :spin:

Can you take a look at it?
It’s here in my google drive: https://drive.google.com/file/d/0B8I7vdW7EC1iZWF6cEpTdVU2bG8/view?usp=sharing

The model is “Varg” and the material is the one with the same name. As you can see I have some stacking of textures but can’t figure out how to bake them.

Thanks!

Most of your maps are 4096x4096 which is going to get huge quick. I’d resize them all to 2048x2048, except for the normal which I think you’re not supposed to change the resolution later. On the fur have the segments lowered to 2, which will help with the particle memory.

And would it help if I baked the textures to one?

And my journey took an unexpected ending: when I at last had baked my textures I FINALLY SUCCEDED TO RENDER IT IN MY GPU!!! Buuuuuuuuuut it turned out that the GPU render subsurface scattering SLOWER than the CPU O.O
Well, not much to do about it… :spin:
Thanks for all :smiley:

You also should probably start breaking-down the shots and doing a lot more work in the compositor.

Give up, early, on the idea that a scene has to pop-out of one render, fully-formed, like Venus popping out of a big clam-shell. It doesn’t have to and it usually doesn’t.

Also give some serious thought to whether you want to or need to use Cycles for everything. It’s a good renderer for nice, soft, bouncy lighting but it can sometimes take a long and inexact way to get there. You have more than one renderer at your disposal, just considering what’s built in, and a great compositor. Compositing will save your ass-terisk… :yes:

Thanks, that’s actually what I choosed to do in the end: I seperated my character model and its fur in two seperate layers and they render very nicely. These render layerss truly are life savers when it comes to rendering…

Yeah, embrace that idea. Stuff like: if the camera’s not moving, you need one background frame. Freely combine BI with Cycles to use the best of both. Turn off shadows, then use shadow-only spots to create shadows where you want them. Separate everything into individual layers so that you can “mix-down” the final shot like they mix-down a song in a recording studio. Put detail in close-up shot #1 and they’ll “see it” in farther-away shot #2. Cut-together all the shots first, using OpenGL, then render exactly what you need. Stuff like that. You wind up planning a lot more in advance, but the shoot takes dramatically less time … and I think the results overall just look better.