assuming vram is 1g, if the rendering goes beyond that, will cycles be able to handle it by using system memory?
I would like to know that as well because the VRAM-Limit is THE disadvantage of every GPU-Renderer.
Cycles has already a very smart UI and Material-Nodesystem.
I guess with Brecht we have the brain to also create a very smart VRAM-Management
No, it won’t. Loading things stored in system memory into vRAM is god-awfully slow and would bog down your renders to the point that it might just be faster to use the CPU. Also, raytracers by nature usually need all scene data loaded to each rendering device at once because of the way that rays interact with surfaces, and it is therefore very difficult to render in buckets as you might do with a scanline renderer. There are people working on developing lazy loading algorithms for the GPU at the moment, but they are all painfully slow, and mostly confined to the academic realm at the moment. The chances of seeing it in something like Cycles any time soon are pretty slim. This is the reason that even the most advanced production path tracers don’t even touch the GPU.