CAD tip # 8529:

Compared to the standard, software-only computing (CPU-based), this rendering is multiple-times faster (e.g. TESLA GPU will be about 6x faster than a modern quad-core CPU). The number of CUDA cores contributes to the computation speed in a linear way (128 cores will be 16x faster than 8 cores). For iRay you can use GPU on the nVidia GeForce, Quadro and Tesla graphic cards. To maximize the rendering performance, set your nVidia driver to SLI and ECC values "Off", and the parameter PhysX to "CPU". If you have a single graphic card in your workstation, response of your interactive rendered scene-viewports will be substantially slower during iRay rendering.

The iRay renderer uses all available GPUs in your PC (and also all CPUs). If you want to limit the usage of GPU units, use e.g. the application iRay Manager [URL verified 3/2012].

In the rendering settings (F10), on the Common tab (Assign Renderer), set "iray Renderer" as the production renderer, and optionally adjust its parameters (the Renderer tab). For lights and environment use the Mental Ray (mr) versions.