I am investigating rendering architectural interiors as stereo equirectangular spherical panoramas for viewing on a GEAR VR headset. I’ve had some great results with simple scenes but when I add, furniture and other props the scenes get too complex for my GPU and rendering on the CPU takes too long.
To keep noise to acceptable limits, I find I need to render scenes with 1000 samples. I have put some time into making sure my glass materials are good, I use portals (though they don’t seem to make much difference) and use clamping.
The best I can manage overnight (on the CPU) is rendered at 250 samples - and the result is really too noisy.
My hardware setup is probably not quite up to the task:
Intel Core2 Quad CPU 2.83 GHz
8 Gb RAM
NVIDIA GTX 660 GPU
I’d be grateful for any advice on how to achieve better results. In particular I have the following questions:
Is it possible to suspend or lower the priority of a rendering task during the day and to resume it again at night ?
I don’t mind waiting a long time for my results, I just can’t afford to lose my PC during the daytime.
Is there a way to break a render task up into smaller groups of tiles and to stitch them together reliably later (using something like photoshop) ?
I note that the worst noise is on large, flat, diffuse surfaces. I have heard that ‘baking’ might be appropriate here but I don’t properly understand how it works. Would it be possible to bake just the room interior on the GPU using a large number of samples and then to add furniture/props etc and render the whole composition using a smaller number of samples on the CPU?
Are there other ways of breaking down a complex rendering into smaller tasks ?
Should I buy new hardware? How much of a performance improvement could I expect ? (That said, I’d always prefer to think my way round a problem before spending my way out of it).
All advice gratefully received.
Thanks in advance