3D Animation and Visual Effects Message Board › Graphics Processor Unit (GPU): Observations on use in a Render Farm setting
|A former member||
You may already know that Graphics Processors (GPU) boards can dramatically speed up render times but do you know when to use a GeForce (such as a GTX-580 or GTX-680) versus a high-end Tesla or Quadro board?
I use GTX-580, GTX-590, GTX-680, Tesla C2070 and Quadro 4000 boards. Gamer type boards such as GTX-580 work really well for rendering on Blender and/or IRAY (special plugin to use GPU for rendering on Cinema 4D). These boards even tend to render faster than Tesla boards; however, high-end Tesla and Quadro boards tend to have more video memory than GeForce boards.
If you are rendering smoke or fire with TurbulenceFD, you need LOTS of video memory on the GPU. I've noticed when rendering smoke it has taken more than 5 Gigabytes of video memory. Geforce boards are typically 1.5GB, 2GB, 3GB or 4GB (GTX-690). I've never heard of a 6GB GeForce board. So...that's where you'll need either a Quadro or Tesla board if you want to go the CUDA route (Nvidia's Compute Unified Device Architecture).
These are just my observations from actual testing. You may also need big video memories on your GPU if you are rendering large textures. Just something to keep in mind when you consider gamer-type graphics boards versus expensive Quadro and Tesla boards.
I haven't experimented with OpenCL or ATI boards (an alternative to the Nvidia CUDA GPU) so don't know how effective they are. Maybe someone with experience here can post?
The above statements are based on tests I've conducted in my renderfarm with these GPU's. Just passing this along in the hopes it may provide some insight and/or help to anyone considering using GPU's themselves.