1. Download UNIGINE Heaven 4.0 from here - Windows / macOS / Linux 

2. Run two Benchmarks with the settings on the picture - one with External Display (Monitor) and one with Internal Display (if you can): 

EXAMPLE 1: AKiTiO Thunder 2 + EVGA GTX 1060 6GB SC / MacBook Pro (Retina, 15-inch, Later 2013) - Thunderbolt 2 @16Gbps

External Display (Monitor):

Internal Display (Optimus):

EXAMPLE 2: AKiTiO Thunder 3 + EVGA GTX 1060 6GB SC / MacBook Pro (Retina, 13-inch, 2016) - Thunderbolt 3 @32Gbps

External Display (Monitor):

Internal Display (Optimus):

To Run Internal Display we have 4 ways based of your laptop:

for Windows if you don't have dGPU -  https://egpu.io/forums/implementation-guides/2013-15-mbpr-iris-only-gtx106016gbps-tb2-akitio-thunder2-win10-internal-display-optimus-enjoy

for Windows if you have dGPU -  https://egpu.io/forums/mac-setup/how-to-keep-mbps-irisiris-pro-activated-when-booting-into-windows-boot-camp/#post-1458

for Windows if you have dGPU but you can't activate Iris (iGPU) -  https://egpu.io/forums/implementation-guides/2012-15-mbpr-gt650m-gtx106010gbps-tb1-akitio-thunder2-win10-enjoy/

for macOS -  https://egpu.io/forums/implementation-guides/mid-2012-macbook-air-gtx106010gbps-tb1-akitio-thunder2-macos-enjoy-internal-display/

AKiTiO Node + EVGA 980Ti Hybrid 6GB / Macbook Pro (Retina, 15-inch, Mid 2015) - Thunderbolt 2 @16Gbps - macOS Sierra 10.12.4

External Display: 

Internal Display: 

Akitio Thunder 3 +  ASUS STRIX GeForce GTX 980 Overclocked 4 GB / Intel Skull NUC (i7- 6770HQ with HyperX Impact HX424S14IBK2/32 32 GB (2x16 GB), 2400 MHz, DDR4, CL14 SODIMM) - Thunderbolt 3 @32Gbps

External Display: 

Internal Display: 

MacBook Pro (15-inch, 2016)
2,7 GHz Intel Core i7
Rog station 2
Evga 1080 Hybrid

I only took with external monitor since it wont run without tricks

R9 270 External@2Gb/s

CPU Power Managment disable, no Turbo.

AKiTiO Node + GTX 1080Ti FE 11GB / MacBook Pro (Retina, 15-inch, Late 2016) – Thunderbolt 3 @32Gbps

External Display (Monitor):

AKiTiO Thunder 3 + MSI GTX 1070 6GB Aero OC / MacBook Pro (Retina, 13-inch i5 touchbar late 2016) – Thunderbolt 3 @32Gbps

External Display (Monitor):

AKiTiO Thunder 2 + Gigabyte GTX1050Ti / Lenovo T430s - Thunderbolt1 @ 8Gbps

External Display:

I cannot run 1080p on the internal display as it is a 1600x900 panel and unlike 3DMark benchmarks, Heaven reverts to 1600x900 if I select a higher resolution (3DMark renders at the selected resolution, then downscales to the display).

Akitio Node + GTX 1080Ti 11GB / MacBook Pro (Retina, 15-inch, Late 2016) – Thunderbolt 3 (Windows)

System

Settings

Here are some 4K benchmarks using a MacBook Pro 2.9GHz, Akitio Thunder3, and Zotac GTX1080 mini. I've placed links underneath to reviewers posting benchmarks from their desktop GTX1080 setups for comparison.

4K|Ultra|8X-AA|Tessellation: extreme

https://www.youtube.com/watch?v=NDu5xovLUO8

4K|Ultra|4x-AA|Tessellation: normal

http://techgage.com/article/nvidia-geforce-gtx-1080-review-a-look-at-4k-ultra-wide-gaming/6/

4K|Ultra|No-AA|Tessellation: extreme

http://www.fudzilla.com/reviews/43048-nvidia-geforce-gtx-1080-ti-review?start=11

4K|Ultra|8X-AA|Tessellation: extreme| Overclocked

Akitio Thunder 2 | Zotac GTX1060 6GB | Mid 2011 27" iMac i7 2600 | 27" AOC External Monitor

 Tested in macOS

Windows Side

Akitio Node | GTX 1070 | Mid-2012 Retina MacBook Pro 15" with 2.6 GHz processor | 27" Dell 1440p external monitor

Mac OS

External screen:

Internal screen:

Windows 10:

External screen:

Internal screen:

Similar to how their is an implementation database and page on here, it would be great to actually get a benchmark database.  Would need information on Windows vs. Mac OS, Machine (e.g. what model), processor type/speed, thunderbolt version, enclosure, video card, internal or external screen, and resolution.  If we had that it would be fascinating to see some of the different results and compare where the bottlenecks occur for different cards and different thunderbolt versions - for instance by scanning through this I'm seeing that my 1070 on TB1 is performing almost as good as a 1070 in TB3 on a 2016 13" MBP (72 vs 66 FPS), and almost as good as a 980Ti in a TB2 MBP (72 vs 66) in Mac OS at 1080p on an external monitor.

I think it would also be super helpful for people when they are thinking about what card to buy, and will help answer the question of is a 1070 worth it in a TB1 computer compared to a 1060 (as far as I can tell, yes) or is the extra power wasted because of the bandwidth limitation.  Also, some people will want to game on their internal screen and will want to know what card to buy to do that at a minimum FPS, others will want 1440p or 4K, etc.

Not sure who would have to create the database and page, but if there's other interest I'd love to see it put together and we could have a simple data input sheet and way to attach screenshots as proof.  What do you think?

@ozzy great idea. We will look into building a database for GPU performance through Thunderbolt enclosures. 

Hp Spectre X360 13 i7-8550U 16GB Ram 
Asus ROG XG Station 2
Nvidia GTX 1080 Ti
Alienware 34inch 4k screen 

FullHD

4K

2016 Apple Macbook pro 13 w/TouchBar

Aorus Gaming Box 1070

2M Cable Matters Thunderbolt 3 Cable

Windows 10 via Bootcamp

Internal Display

Unigine Heaven Benchmark 4.0

System

Settings

Hey folks,

I have some very interesting results to share. If you look at my activity, you'll see I have a 2015 iMac 4k (iGPU only Iris Pro 6200 Graphics). The tl;dr summary is that I could not find a way to engage Nvidia Optimus. The eGPU does not see the internal iMac display. The only way I could find to engage the eGPU was to use a headless HDMI adapter and then use the Dual Display option in Windows Display Settings. There are some exceptions - some applications will engage the eGPU without the HDMI adapter plugged in, such as Fortnite. Others won't. The only 100% way to use the eGPU resources is to run in Dual Display mode. 

If you want to take a look at my hardware info, check out my setup on Userbenchmark.

Onto the results, then something very interesting at the bottom ;-).

Dual Display Mode (1o8op) with Headless HDMI attached

Dual Display Mode (1o8op) with Headless HDMI attached (Also Recording using GeForce Experience overlay)

1080p eGPU Internal Display - Unplugged Headless HDMI adapter while application was running

Dual Display Mode (1o8op) with Headless HDMI Retest - Plugged HDMI adapter back in while application was still running

3d Performance In Task Manager While Heaven application running - 1080p Dual Display Mode using HDMI adapter

3d Performance In Task Manager While Heaven application running - 1080p Internal Display only (unplugged HDMI adapter while application was open)

The most telling information is in the last two images: When running in Duplicate Display mode, the eGPU is running at 100% while the iGPU is running at 20%. When rendering without the HDMI adapter plugged into the eGPU, the eGPU shows 66% and the iGPU 14%. I would assume this is why the Dual Display benchmark shows better performance versus single display mode.

As to why this is happening, I have no idea! Does it have something to do with the Intel 3d Graphics settings? Is there a way to get Nvidia Optimus working as designed on my system (doubtful)?

Edit to add CUDA-Z information: