OTOY Forums

Does octane performance depend on the pci connection to the CPU speed while rendering?

Or, once the scene is loaded in GPU memory, is the card somewhat independent of the rest of the computer while rendering?

The reason I am asking is with some PCI expansion boxes you can run multiple x16 slots....but the connection to a motherboard is done through a single x16 slot...and the cards in the expansion are all sharing a max of 16 lanes when communicating with the CPU.
Would there be a big performance hit in this case (some boards are up to 20 x16 slots...all connected through 16 pci lanes to the CPU)...or once the scene is loaded into memory...is the CUDA rendering mostly self contained.

There would obviously be a small hit in scene loading time...and some interactivity lag...but more interested in final render time.

For rendering the GPU runs independent of the CPU. The main bottlenecks are loading the geometry and textures onto the device, and, when rendering big images, loading the rendered image from the device. When changing material or rendering settings usually there is only a small transfer needed, so these updates should still be fast.

--
Roeland

thanks