Вообще вопрос был про скорость конечного рендера.
Та вот в доке по Mental Ray сказано:
By default, mental ray uses a scanline rendering algorithm for primary rays when no lens shaders are present that modify the ray direction. This algorithm can cope with both static and motion blurred scenes. In addition, mental ray supports OpenGL®1.4 hardware to further accelerate scanline rendering of static scenes (without motion blurring). If the master host provides OpenGL acceleration with sufficient resolution, mental ray can use this to generate acceleration data that is subsequently used during regular rendering. This combines the speed of a hardware OpenGL accelerator with the full shading capabilities of mental ray, and is particularly effective for scenes with high polygon counts.
Shadow maps can also be rendered with OpenGL acceleration. This is particularly effective in mental ray 2.1 since shadow maps need no shading. Most of the computation involved is then performed by the OpenGL system, which greatly improves performance. mental ray 3.0 renders only small portions of shadow maps on demand, which in some cases is faster than letting OpenGL render the entire shadow map at once.
However, the accuracy of OpenGL acceleration is generally slightly lower than that of the standard software scanline rendering algorithm. Also, OpenGL acceleration is limited by the allowable OpenGL image resolution.
Так что, как видим, с оговорками но влияет. OGL или ещё что ускоряет скорость рендера faza не интересовало. Вывод хороший акселератор способен ускорть скорость конечного просчёта сцены.
Повторюсь на машине P4 @ 2500 Mhz + 512 mb Rambus @ 533 Mhz разница между картами Matrox G400 и ATI Radeon 9000 pro составляет процентов 10-30
