Two new eGPU solutions on macOS 10.13.4: pure EFI and hybrid  

v1.0.0 - 1.0.4 do not support any 2018 MBPs or iMac Pro due to T2 chip's encryption (even if FileVault is turned off). You have to wait for v1.0.5.

Instructions in short:

1) Follow the steps of the installer
2) Check that FileVault is disabled
3) Select "EFI Boot" by holding down the Option key on startup

Latest version 1.0.4

https://egpu.io/wp-content/uploads/2018/07/automate-eGPU-EFI-v.1.0.4.pkg_.zip

- added support for all TB1 Macs
- dropped support for MacBookPro 11,3 (also other Nvidia dGPU equipped models not recommended, please use 1.0.3)
- AMD dGPU + Nvidia eGPU framebuffer conflict solved

This special version introduces three new Hybrid modes:

1) Hybrid (iGPU)
2) Hybrid (AMD dGPU)
3) Hybrid (closed-clamshell) -> black internal screen, neither iGPU nor AMD dGPU show "Color LCD" under Graphics/Displays

You should always hot plug the eGPU when using any of these Hybrid modes, AFTER logged into macOS. The app logs you out automatically and enables eGPU screen output.

More info:
https://egpu.io/forums/mac-setup/two-new-egpu-solutions-on-macos-10-13-4-pure-efi-and-hybrid/paged/9/#post-39650

What's new in 1.0.3
- fixed "Invalid Nvidia kext found" error message if web drivers are not installed (1.0.2 didn't check the kext existence)
https://cdn.egpu.io/wp-content/uploads/2018/07/automate-eGPU-EFI-v1.0.3.pkg_.zip

What's new in 1.0.2
- added support for GeForce GTX 780 Ti
- SkipSetupDataForOSX=false setting now works (primarily meant to be used with "Pure EFI" if you want eGPU screen output)
- deletes unnecessary automate-eGPU.kext from the install location

https://cdn.egpu.io/wp-content/uploads/2018/07/automate-eGPU-EFI-v1.0.2.pkg_.zip

UPDATE: 01.07.2018 automate-eGPU EFI v1.0.1 released:  https://egpu.io/wp-content/uploads/2018/07/automate-eGPU-EFI-v1.0.1.pkg_.zip

What's new in 1.0.1

- added support for GeForce GTX 780 3GB, GeForce GTX Titan, GeForce GTX 780 6GB, GeForce GTX 750 Ti, GeForce GTX 1060 3GB, Quadro P2000 5GB
- ValidateNvidiaKexts() and ValidateOtherKexts() in the preinstall stage (modified kexts not allowed)
- config.plist ForceIntegrated changed to MuxSetting [0, (do nothing), 1 (integrated), 2 (dedicated)]
- AMD dGPU equipped MBPs have MuxSetting 1 automatically
- DefaultVolume EFI logic refactored
- The installer asks to disconnect the eGPU enclosure to avoid kernel panics during cache rebuilding
- config.plist CsrActiveConfig value is now 0x67 (disabled)

Known issues with this pkg: Pure EFI doesn't produce eGPU screen output due to simplified boot.efi chain loading

UPDATE: 23.06.2018 automate-eGPU EFI v1.0.0 released:  https://egpu.io/wp-content/uploads/2018/06/automate-eGPU.pkg_.zip
More info:  https://egpu.io/forums/mac-setup/two-new-egpu-solutions-on-macos-10-13-4-pure-efi-and-hybrid/paged/2/#post-38566

---
The most observant readers may have noticed my recent investigation on the EFI side, making possible to use both Nvidia and AMD cards out of the box, without needing to backup kexts because Apple's system files stay untouched.

EFI approach provides more reliable Nvidia eGPU screen output but lacks hot plugging support because it bypasses Thunderbolt layers. The eGPU is detected as if it was installed internally. See the screenshots below.

The "Disconnect" menu bar item is not visible. What happens when we combine EFI with our earlier discoveries? The EFI and yifanlu's kext codeless automate-eGPU.kext operating in "hybrid" mode?'' Boot into macOS, hot plug the eGPU on the login screen, and boom:

Multiple eGPU support might be a bonus side effect. The "Type" column says "External GPU". This is all done without modifying system files. The EFI is executed from the USB stick in my test scenario, but it can be run from any media.

Does it work with AMD cards through TB2? Yes.

Confirmed to work with the nMP, Late 2014 Mac mini, Mid 2015 15" MBP (M370X), and Late 2013 13" MBP.

The EFI does it on the fly. No magic, it will be all open source. My next goal is to set the iGPU as the primary device, resolving the sluggish AMD dGPU and Nvidia eGPU cooperation. It seems to work manually, but we want automation. Stay tuned.