Someone recently pointed CUDA out to me and I wanted to know of any success stories of building a small render farm of pc’s with nvidia 8.x cards in them and using them either with mental ray (to render ambient occlusion maps), building lightmaps, or compressing textures for shaders down into the proper format.
We are looking at starting small with 4 boxes with sli which gives us about 8 8.x cards.
Anyone have any stats on how much faster such processes in theory are than just using like quad core cpu’s? I know the quad core cpu’s are more expensive, but I just wanted to get any feedback from the community here before digging into the api and working on the code.
For anyone who has done something similar, would you suggest using windows or the linux platform (pro’s or con’s?)
does anyone know if their are any plugins for maya that use mental ray or turtle that instead of using cpu satellite’s can use the gpu’s on the render farm?
mental ray won’t use the gpu for rendering occlusion passes. there is some gpu support in mental ray (not CUDA-based though), but it is very limited and practically not used.
Just for myself to check if I am not mistaken, I thought that you had to disable SLI in the driver right?
(Just because I am looking at the 9800GX2 as a possibility)
I think you also need to de-bridge it. So pull of the bridge. I know this is a bit hard with the 9800X2 I love to see people solder the new stuff tho… :P
But I’m not familiar with disabling it at driver level.