Rendering Workstation

We get several requests like this each week, and they are very difficult to answer well. It's not that we can't help you, but the time involved to do it correctly is extensive.

Some day I'll get a workstation guide in place, but even that will not do the work for you.

There is only one way to go about this, and that is by doing the research yourself.

Questions to address:
1. Where do you anticipate the majority of time will be spent? Which program? List out the programs in order of anticipated usage.

2. How are these programs designed? Are they CPU intensive or do they need a strong GPU? The TYPE OF WORK you will do with these programs may also be a factor. Certain types of work in MAYA, for instance, may not require a special GPU at all.
One program may need and use lots of CPU cores... so much that two quad core hyper-threaded Xeons are in order. Other programs may be reliant on the CPU, but are not written to make use of many CPU cores. For these types of programs, the CPU needs to be very fast, but does not need many cores.
Another program may do much of it's processing with the GPU, but which kind? Does it need a professional GPU like the Quadro? Or will it do just fine with an enthusiast grade card?

So you do some searches. For instance "maya hyper-threading" "after affects GPU"
"Premier Pro CPU benchmarks".
You end up at forums where these specific programs are discussed. You take all that into account and then you come back to this thread with certainty about your CPU and GPU choices... and I'll help you with the rest  

And set a budget.

Thankyou.

No use in getting the Quadro unless you're doing a lot of serious high polygon count work or your application specifically requires it. IF you're going to be using GPU-based rendering, iRay for example, you can benefit from having a second video card only while rendering... otherwise when it comes down to render time, you're better off having another machine networked to your primary system.

Draven35:

Just confused when you say GPU "rendering" you are talking about GPU accelerated preview right ?
I believe final o/p is rendered on the cpu or it is CPU "intensive"
need to be clear the confusion on it.!

No, I mean rendering of final output on the GPU. iRay, the GPU-based version of Mental Ray, can use GPU, CPU or both to render final output.

If you're doing extensive CPU-based rendering. once again, you're going to need a render node far far more than you'd need a Quadro. 'Needing a Quadro' is also hinged on whether you will even stress the GPU enough to see the difference between a Quadro and a GTX.

Ye agreed,
Iray Vray Mray ..and Blender and Octane and so many more I have been doing this all day trying to figure some decent way of FINAL Gpu rendering , The 3D work is not very intensive that would saturate Quadro , My priority is Faster Render times! and GPU is far more efficient thn CPU due to CUDA , if utilized properly there are some render engines out there which utilize GPU to max for Final output , even SLI support.
I have no experience with these new render engines for GPUs , so i am goin to do more research and see if I can utilize the GPU to max anytime soon!

iray = mental ray on GPU

VrayRT= VRay on GPU

Same old renderers, they just run on the GPU.

Note that neither is necessarily 'real time rendering', and they are not likely to be using the graphics card to render output in the same way that they render previews.

A Quadro is the way to go..if u can afford it get a quadro and a tesla c2000 series card (c2050, c2075, c2090)...by having these two cards working together it offloads the rendering to these cards..they will more then cut render time in half

If you're using a GPU based renderer (like iRay) then yes you can spread the render across multiple cards, even dissimilar cards. You could even have a Quadro as your primary display card and a GTX-series consumer card (which is faster ad has more CUDA cores) as your secondary card and farm the renderings out to it.

Keep in mind the following:

1: you need to be able to load your entire scene into the GPU's memory. If it doesn't fit in the graphics card's memory, you can't use GPU-based rendering.
2: (note for anyone else reading thread) iRay and VRayRT both only support NVidia GPUs.
3: Both renderers have limitations- many shaders for the software renderers do not work in the GPU-based versions. This includes subsurface scattering (essential for characters)