When I check the GPU-option in 2.73 and try to render a project in Cycles, I get this message:
Mem:26182.43M, Peak 26182.43M - Cancel - CUDA Error: Out of memory in cuMemAlloc(&device_pointer, size)
When I use my CPU instead, the image is rendered in 5 seconds.
Here are my specs: i7-4820K 3.70GHz, 16 GB RAM, Win 8.1., 2 x GTX Titan, latest Win/Titan drivers.
(I did see the “GPU Rendering with Cycles”-thread, but the systemagnostic-link is dead.)
Your peak memory is 26GB. That’s far larger than ANY GPU could possibly render. When rendering with GPU you are limited to the available vRAM of the smallest card.
Thank you for your reply. But isn’t it a bit odd that my CPU can render the project in a few seconds, then?
For your information, I get faster, less noisy images with 2.70 then with 2.73.
Also, have you tried rendering different layers at a time? Maybe some objects that don’t reflect/get reflected could be moved to a different layer and could be rendered with your CPU simultaneously (with multiple blender’s open at a time) or rendered at a different time. I read it somewhere, can’t find where though.