A GTX 1080/1080 Ti in a AORUS/Gigabyte Gaming Box/Sapphire GearBox/AKiTiO Node, on macOS High Sierra
First and foremost: what card is best for you will depend heavily on your specific workload, down to often even relatively small details. Things from what you're modeling, render settings, and even what format you're exporting to can influence what card works best for you massively. Unfortunately, spending more doesn't mean you'll be getting better performance, and outside of the very best Quadros, doesn't even guarantee the same level of performance. What I'm trying to do here is provide you with a general recommendation of what will probably be good given your (somewhat varied) workload. If it turns out that you're really doing one thing in particular, either come back here for more specific recommendations or contact the developer/a third party consultant who specializes in the tech for a specific recommendation.
Second: I am by no means an expert when it comes to hardware-accelerated GPGPU. The world of GPUs, even compared to others in the industry, is a particularly messy one, and I make no pretense that I understand it all or even close to it. If you want a brief and very interesting (while slightly terrifying) look into the world of graphics, I highly recommend this post on the Dolphin Emulator blog on ubershaders.
That said, let's get into it.
Understanding what GPU is best for you requires understanding how GPUs interface with software. Based on your background, I'll assume you understand the basic premise between APIs, but it's worth reviewing them in specific for macOS, since it in particular sits in a bit of a weird space when it comes to graphics APIs. For the purposes of these programs, macOS has three different graphics APIs: OpenCL, CUDA, and Metal. OpenCL is cross-vendor and cross-platform, working on AMD, Nvidia, and Intel (plus others) GPUs with Apple's out-of-the-box drivers on macOS. CUDA is specific to Nvidia GPUs but cross-platform, installable with Nvidia's web drivers on macOS. Metal is the newest of the bunch, written by Apple for their OSs with support for AMD, Nvidia, and Intel (plus others) GPUs. So, which to choose? Well, it's not so simple. Metal and CUDA tend to provide "better" access to the hardware, at the cost of the developers. Although Metal might seem like the obvious first choice, because it's so new and specific to Apple's OSs, performance is often somewhat lackluster, if developer support exists at all. OpenCL might seem like a solid bet, but because Apple wants to push Metal, it refuses to provide drivers past version 1.2, even outright deprecating it with the release of macOS 10.14 Mojave (somewhat ironic given Apple is the original author of OpenCL.) Additionally, Nvidia GPUs tend to perform worse with OpenCL compared to CUDA. CUDA, meanwhile, suffers from the obvious burden of only supporting Nvidia GPUs and lackluster support from Nvidia themselves.
This has left software vendors in a bit of a tight spot. Many have existing support for OpenCL (and sometimes CUDA) based hardware acceleration, but don't see a clear path forward. No matter what option they choose, they're either alienating part of their consumer base on the basis of their hardware choices, getting less performance than they could, or spending more time and resources developing their software. As a result, developers have basically scattered when it comes to what graphics API they use on macOS.
For this reason, this will be considered on a program by program basis, along with hardware.
After Effects (and sort of the rest of CC)
In the timeline, After Effects supports some effects and fast rendering on GPU using OpenGL for most, and CUDA only for a few.
As it stands, Media Encoder still favors OpenCL over Metal for encoding. Or maybe not? Initial testing with Adobe Media Encoder 2019 (which has a bunch of improvements on the Metal front) actually yielded a 11% (1:20 vs. 1:30) improvement with Metal transcoding from a H.264 screen recording to H.264 Youtube 1080p... so maybe not? In any case, I won't be considering this for now because of the wonky state of Metal at the moment (more on this later.)
Let's get down to the benchmarks:
Conclusion? More raw power is better (but not by a ton); Nvidia is favored somewhat over AMD; "Pro" features don't matter.
Unfortunately, numbers for the RTX series aren't available here. However, we can still get a pretty good idea of things.
AFAIK, Maya doesn't support GPU-based rendering. While I believe there are other renderers which do, I don't feel I'm qualified to talk about them here.
Conclusion? More raw power is better; AMD and Nvidia appear roughly equal; "Pro" features are favored by ~5-20%, but which features are important is unclear.
Unfortunately, SPECviewperf numbers aren't available for Cinema 4D AFAIK, and I haven't been able to find a good replacement. Will update if I find something.
Cinema 4D is really simple: no support for GPU rendering!
...until it's not. While the "normal" rendering engine doesn't support GPU acceleration, Cinema 4D offers the ability to use a variety of other rendering engines, including Arnold, Octane, Redshift, and even Blender's Cycles engine. However, these might as well be considered separate programs, so I won't actually be covering those here.
Except, starting with release 20 on September 1st, 2018, Cinema 4D includes built-in support for the Radeon ProRender rendering engine, with support for OpenCL and Metal based acceleration. Unfortunately, because it's new, numbers are hard to come by. So far, all I've managed to find is this set:
Note also that these numbers are with OpenCL on Windows 10 Pro and don't include the RTX series.
Conclusion? More raw power is better; AMD is favored over Nvidia; "Pro" features don't matter. However, things are fast-moving.
GPU: The final decision
Based on your budget, ~$500 seems reasonable for your graphics card.
Frankly, the extra expense of a "pro" card doesn't make sense for the relatively minimal gains in Maya. Because of the much better availability of consumer cards, I'd highly recommend one of them over a pro card.
Going for straight power, a used GTX 1080 should be well within reach; and, if you're willing to shop around a bit, a GTX 1080 Ti should be doable. Right now, AMD's offerings just aren't competitive at this price point; used 10 series cards are far cheaper and available compared to Vega cards, and no AMD card can match the 1080 Ti ATM.
However, it must be noted that Nvidia cards are currently unsupported on Mojave, and other than Kepler, Metal drivers are absent, making it impossible to upgrade to Mojave. However, it seems likely that Nvidia will release drivers (with Metal support) in the future. Just when exactly that future is is unclear. In the meantime, stay on High Sierra.
Note also that Thunderbolt 2 and Nvidia eGPUs are unsupported by Apple and require extra work for macOS compared to the plug-and-play of an AMD card on Mojave with a TB3 equipped Mac. Since you've got a Thunderbolt 2 equipped laptop anyways, I'm going to recommend this setup anyways. Just keep this in mind.
eGPU.io is a fantastic resource for everything eGPU related, from hardware recommendations to build guides. Information is generally quite recent, and the forums there are extremely useful.
I'm going to recommend using a Thunderbolt 3 eGPU enclosure with an adapter, rather than a Thunderbolt 2 enclosure. While Thunderbolt 2 enclosures can be cheaper, they're usually harder to find, have much, much lower power limits, and aren't great for upgradability.
eGPUs are a bit strange in that there are a bunch of different enclosures available, all of which are fairly valid choices. While I'm going to make some recommendations, if you find yourself looking at something else, there's nothing wrong with that. Again, the eGPU.io buyer's guide is a fantastic resource.
This box is a pretty good deal at $670... if you can find one. Unfortunately, as far as I can find, they're completely out of stock in normal retail channels. However, they're still in good supply from 3rd party resellers for around $600 with the card, and often significantly less without the card.
They're both very similar budget enclosures, each with a few trade-offs. Winner of each category gets asterixes:
| GearBox | Node
PSU type-location | fATX-int | *SFX-int*
Power delivery (PD) | *60W* | 15W
IO | *2x USB 3.0, 1x Ethernet* | None
Frankly, most of these features are probably unimportant given your laptop and the already beefy PSUs included out of the box. So, it's really a toss-up IMO.
It's bigger, has a full sized PSU, and looks prettier. It also costs more. IMO, nothing more than a runner-up if you think the others are too ugly.