Here's a build focused on maximizing export render performance and scalability for CPU based rendering in Keyshot (Zbrush default), Vray (plugin for autodesk apps), and Mental Ray (Autodesk Built-in renderer). It will also provide excellent performance in the C4D ray-tracing engine, which is now a plugin for After Effects. And of course, all this CPU strength will also scale nicely with video editing in premier. The following build is not intended for use with any GPU accelerated export renderer's like Furyball, V-Ray-RT, Octane, I-Ray, etc.
CPU: E5-2640 V3 $900
GPU: FirePro W7100 8GB $650
HSF: SilverStone AR01 $35
MOBO: Supermicro X10DAI $390
RAM: 4 x 8GB CT4K8G4RFS4213 $390
SSD: Sandisk X210 512GB $250
Storage: WD2000F9YZ $125
PSU: Seasonic SSR-750RM $120
CASE: Phanteks Enthoo Pro $90
~$3000 (no software)
Depending on your workflow, you may be able to make do with a GeForce GPU, though there are viewport options in these 3D creation applications (most especially in Maya) that are only well supported, and/or only perform well on a workstation GPU. The W7100 strikes me as potentially the best choice here for balance between viewport performance and support (especially with its 8GB VRAM buffer), and gaming performance (Tonga core). Though it may be worth trying a gaming GPU for awhile, with the option to upgrade to a workstation GPU later if the gaming GPU presents problems in viewports (Gamble). Most AutoDesk programs are being migrated to viewport 2.0 with the option to run DX11. If I were personally forced to choose a gaming GPU for this workstation, it would be a GeForce, as GeForce cards tend to pass AutoDesk qualification tests more thoroughly than Radeon (some even make the recommended/certified lists), and Nvidia's DX11 implementation on gaming GPU's has better API/driver level optimizations, which would be preferable in compute bound DX11 viewport conditions. So, yea, if you want to save a few hundred bucks, you could take a gamble on a GTX960 or GTX970 instead.
Another possible tradeoff to consider, is compromising on the GTX960 so that you can afford to start the build off with an E5-2660 V3 CPU instead, (25% faster than E5-2640 V3).
If it isn't obvious already, this build proposal is on a dual socket motherboard, but I'm advising the build start out with a single CPU, and have the option to double export render speed with the addition of another equal CPU at a later date. Dual socket boards offer the best compute density and flexibility for combined workstation and render machines.
----------
I think it's worth pointing out, that all of the export renderers you are likely to be using, not only scale well across many cores, and even across multiple CPU sockets, they also scale well to cluster computing configurations over networks, so there is another approach to this build using consumer hardware (rather than enterprise hardware) that I would like to share as an alternative. The tradeoff, is a reduction in compute density/efficiency and increased configuration complexity (have to build and configure multiple machines, render servers, etc). The advantage is more performance for the money.
Workstation/Gaming machine:
CPU: i7-4790K ~$330
HSF: Rajintek Aidos ~$23
MOBO: Asus H97M-PLUS ~$100
RAM: 4 X 8GB DDR3 (2 X 997069S ) ~$200
GPU: W7100 $650
SSD: Crucial MX200 512GB ~$200
Storage: WD2000F9YZ ~$125
PSU: Seasonic SSR-550RM ~$75
Case: CM N200 ~$45
~$1750 (no software)
Render Nodes:
CPU: FX-8310 (tiger direct) $120
HSF: AMD OE AM2/AM3 125/140W HSF (ebay) ~$12
MOBO: GA-78LMT-USB3 ~$60
RAM: 2x4GB Crucial Ballistix Sport ~$60
HDD: WD2502ABYS ~$22
PSU: SS-350ET ~$40
Case: CM N200 ~$45
~$360 each. (no software)
So you could do the workstation + 2 render nodes for a bit less than the cost of the Xeon workstation. The result would be up to ~30% better viewport and gaming performance (from raw per-core performance), and up to 80% better export render performance (with all 3 machines rendering).
For roughly the price of adding another E5-2640 V3, we could add 2 more render nodes. In that comparison, the 5 machines combined would still offer up to 30% better viewport and gaming performance, and approximately ~40% better export render performance over the dual socket machine.
There are going to be higher software implementation costs with the render node option, as you may need windows on the render nodes (though many export renderers have linux standalone applications available). Depending on the additional software costs involved, using render nodes may or may not be beneficial. You'll have to do some cost analysis there for the renderers you intend to use.
-----------
There are lots of ways to skin this cat, and the build ideas proposed above may not even be relevant to your intended use if you wanted to be able to support any CUDA based export rendering, but maybe it will get the gears turning a bit.