Originally Posted by jomama22
As stated in the conclusion. The reason pro cards are more expensive and may perform less then their consumer gpus is for the correct bytes of data. Anything relying on precise data needs that exactness of output.
Not exactly. Most Quadro or FirePro cards lack ECC Ram and are thus 99% identical to their gaming siblings as far as hardware goes.
There is no crazy variation that makes the "professional" workstation card compute something more precisely or faster on hardware level.
The differences are in BIOS / Drivers. Workstation cards get "optimized" for accelerating viewports in 3D graphic applications through their drivers, which much like gaming drivers, provide the hardware with "shortcuts" in computing very complex scenes. While games usually feature a few million polygons, medium sized textures and require relatively high FPS, demanding 3D professionals are dealing with scenes that contain tenths or even hundreds of millions of polygons, multiple very high resolution textures, but in general are satisfied with sustaining 15-25 FPS.
In such scenarios, the by comparison "overpowered" GTX/Radeon cards don't have the drivers to help them "cheat" through the process taking shortcuts, but in contrary try to brute-force through the calculations, getting "chocked" by the enormous poly counts / high tessellation of objects. Thus GTX Kepler monsters with 6-7 SMX units might be spectacularly slower than vanilla Quadros with 2-3 SMX units, basically hardware that is identical to mid-range GT cards.
Again, it is not about "high precision" in 3D Apps like Maya or 3DS. You are just after "workable" conditions, just like you are after fast response / playable conditions in games.
In compute scenarios like 3D rendering, each pixel is ray-traced hundreds of times. Thus a few mistakes here and there are averaged out and disappear in the interpolation of neighboring pixels.
Gaming cards work great as 3D rendering GPGPU solutions, as drivers are nearly irrelevant for a properly coded CUDA or OpenCL GPGPU application that almost "talks" to the GPU directly.
Improper OpenCL implementations, like that of VRay RT GPU @ OpenCL mode struggle with AMD cards, despite the vastly superior compute performance GCN cards have, simply because are not OpenCL optimized.
More over, I read half truths about CUDA and OpenCL alike being soooooo prevailing and needed, or rendering modules for programs being some niche 3rd party GPU rendererers like Octane etc, which is simply not true. Certain portions of certain programs are GPGPU capable, and GPU renderers especially are far from mainstream, lacking many of the features fully blown VRay / Mental Ray etc renderers offer. Most CG artists don't have the resources to write their on GPU accelerated raytracers as the CG teams in movies like Avatar do.