Late 2018 Mac mini i3 + RX Vega [email protected] (Sapphire Gearbox) + macOS 10.14.2 & Win10 1809 [theitsage]
I exchanged the 2018 Mac mini i5 configuration for the i3 base model. Given how hot these 8th gen CPUs run, the quad-core i3 seems like a decent pairing with the Mac mini cooling system. First thing I did was to upgrade the stock 8GB RAM to 32GB. While it was apart, I was going to reapply the thermal paste but thought I should run it as is first. I'll do it in a couple weeks then retest thermal performance with new paste.
Late 2018 Mac mini – i3-8100B/UHD Graphics 630 iGPU/32GB RAM/128GB SSD
This Sapphire Gearbox + RX Vega 56 pairing have native eGPU support in 10.14.2 so it was plug-and-play with the 2018 Mac mini. As you see in the photo above, the sound meter showed 48 dB. This was during idle. The noise source was the Sapphire Gearbox 40mm PSU cooling fans. Even in sleep mode these two tiny fan continued running. So did the blue light bar in front. To prepare for Windows Boot Camp, I added a third partition (32MB FAT) for automate-eGPU EFI.
This enclosure has three expansion ports: one Ethernet and two USB 3.1 gen 1. I connected a Matias USB keyboard as Logitech wireless mouse to one of the expansion USB ports. I didn't experience any lagging issues using the mouse in this configuration. This is likely due to firmware tuning. The Sapphire Gearbox has H2D running at half bandwidth to ensure expansion ports stability.
In Windows Boot Camp, there are a couple of options to get the RX Vega 56 eGPU going. The simple method is to hot-plug the AMD eGPU to install Radeon drivers. Once this step is done, switch the monitor connection from the Mac mini to the eGPU. You won't have boot screen and the AMD eGPU disables the iGPU using this method. I wanted to keep the iGPU activated so I'm using @goalque's eGPU-automate EFI as well as disablement of a couple PCIe components. These are the basic steps I took to get this setup going:
- Boot into Recovery mode to disable SIP through Terminal. Open Startup Security Utility to set Secure Boot to No Security and External Boot to Allow booting from external media [in Option Boot].
- Download and install Windows 10 ISO through Bootcamp Assistant [in macOS].
- Download and copy @goalque's automate-eGPU EFI to USB thumb drive [in macOS].
- Hold OPTION at boot to select EFI drive and hit Q to access automate-eGPU EFI [boot selection].
- Select View by Connection in Device Manager then locate and disable PCIe Root Port #2 - A339 and PCIe Controller (x16) - 1901 [in Windows].
- Shut Mac mini down then connect eGPU to Thunderbolt 3 port closest to the HDMI port.
- Boot into Windows using automate-eGPU EFI [boot selection].
- Confirm a new Microsoft Basic Display Adapter in Device Manager then install Radeon drivers [in Windows].
I installed Intel Thunderbolt Software version [220.127.116.110] to confirm the dual Titan Ridge Thunderbolt 3 controllers and their ports. As seen below, Apple set Security Level to No Securiy (SL0). When the RX Vega 56 was working in Windows, I hot-plugged a Thunderbolt 3 to dual DisplayPort adapter to see whether the system would crash. There was no BSOD and both TB3 devices showed up in Intel Thunderbolt Attached list.
CL!ng, Luxmark and Unigine in macOS.
|UHD Graphics 630 GPU||RX Vega 56 eGPU||iGPU and eGPU|
|UHD Graphics 630 iGPU||RX Vega 56 eGPU Internal Display||RX Vega 56 eGPU External Monitor|
In Windows Boot Camp, I installed and ran Steam games on an external SSD connected to the Sapphire Gearbox USB port. There was no lagging issue with the mouse when during an half-hour Uruk-slaying session.
|2018 Mac mini i3||UHD 630 iGPU FHD||Vega 56 eGPU FHD||Vega 56 eGPU QHD||Vega 56 eGPU 4K||Vega 56 eGPU 5K|
|Tomb Raider 2013||8.4 FPS||126.0 FPS||94.4 FPS||49.9 FPS||29.2 FPS|
|Shadow of Mordor||9.3 FPS||102.8 FPS||84.3 FPS||49.5 FPS||30.7 FPS|
|Dirt Rally||12.5 FPS||58.3 FPS||51.3 FPS||39.3 FPS||34.4 FPS|
|Hitman||13.9 FPS||80.3 FPS||76.9 FPS||59.6 FPS||39.0 FPS|
The Mac mini i3 base configuration is a nice eGPU host with direct Thunderbolt 3 connection to the quad-core CPU. Higher specs i5 and i7 configurations should perform better but thermal management is a concern. During the several hours I ran Windows last night, the i3-8100b never went above 91 degrees Celsius. The i5-8500B Mac mini I tested would reach 100 degrees within minutes.
Hello, @theitsage, thank you for posting your findings! I have a question that is irrelevant from your post, but any help would be very much appreciated!
I have a 2018 Mac Mini 1TB/i7/32GB/10G on Mojave 14.0, and the problem I am facing is that the very left thunderbolt 3 port (next to the HDMI) seems to improperly disconnect my OWC eGPU when I even every so slightly wiggle the cable. HOWEVER, this problem does not occur on the other thunderbolt 3 ports, AND when I plug in my thunderbolt 3 to tb 2 adapter into the mac mini's problematic port for my audio interface, I can wiggle the cable and it won't disconnect at all!
Could you (or anyone in this case) help me in seeing if your eGPU disconnects on your mac mini using the same TB3 port that is causing me issues? Thank you!
Picking mine up today but you’ve made me think going i7 may have been a mistake if cooling really is an issue.
Hopefully my set set up will be as straight forward as yours.
Would love to see the same games benchmarked on the i7 too...
Hmmm may have spoke too soon.... worked one and now nope. Hot plugging isn’t even working.
@eightarmedpet I observed some inconsistencies in the way the Titan Ridge controllers detect Thunderbolt 3 devices. I had a fairly easy time getting the 2018 Mac mini i5 going with a number of different workarounds. The Mac mini i3 is more picky. What I learned is Intel Thunderbolt Software can help kick-start an active Thunderbolt 3 connection at boot. I use version [18.104.22.1680] on 2018 Macs. Once you get a hang of which port and how to connect with a particular eGPU, you can uninstall Intel Thunderbolt Software. The eGPU is non-detectable sometimes when using @goalque's automate-eGPU EFI in Bootcamp. I will try the older workaround using rEFInd bootloader and report back.
@parutan Please take screen capture of Device Manager with all PCIe roots expanded. I have not seen one from a 10G 2018 Mac mini configuration. My guess is that it uses the PCIe x16 Controller. I don't know why it would cause stability issues on one of the four Thunderbolt 3 ports however. I used the TB3 ports closest to HDMI port because it's the one I had the most consistent success with different eGPU testing. You can use the port next to it which is on the same TB3 controller.
Yeah think I might just return my Mini. Might write up my experience as got it working again but it’s not great with... guess what? My 5k display.
@parutan I've got an OWC Helios with a Radeon RX 580, similar build to yours but 1G not 10G... struggling to get things working. Somehow got the drivers installed via hot plugging but now seeing error 12 no matter which TB3 port I use. Any pointers would be greatly appreciated.
Ok, I got it working again, but in the most ridiculous convoluted way, and due to that I'm returning the Mini.
The way I got the eGPU detected without error 12 was...
- Boot into Windows normally, using native boot selector.
- Plug in eGPU (error 12).
- Restart (not shut down) with everything plugged in.
- Boot into EFI Bootloader and select Windows.
- Windows boots and eGPU works but resolution is now 4K not 5K (assume something needs to be edited somewhere in the EFI folder which I may look into).
That's too bad about your setup and LG 5K issue. I tried rEFInd yesterday on the 2018 Mac mini to determine whether TB3 eGPU detection inconsistency issue was due to automate-eGPU EFI. The 2018 Mac mini experienced the same flaky behavior too. My best guess is the Titan Ridge controllers and Apple T2 chip have something to do with this. I have seen the same behavior on the 2018 13" MacBook Pro as well. 2016 and 2017 MacBook Pro worked with rEFInd and automate-eGPU EFI much more predictable.
That is frustrating... I'm just back from the Apple store after returning my mini, such a shame as it was nearly the perfect eGPU host.
Back to the 2017 13 MBP!
There are always compromises with choosing a Mac it seems. I like the 2018 13" quad-core MacBook Pro a lot. I don't appreciate the Apple T2 chip being the gatekeeper though. There are instances when one failed boot attempt led to automatic Recovery boot to reset the password.
The 2017 13" MacBook Pro is still a very capable machine. Recently I found a nice deal on an i7/16GB/1TB configuration. I'll run eGPU tests through it soon and make comparisons with the 2018.
That would be super interesting, I only had a chance to test a couple of games and found the fps increase a lot less than expected, I assume performance is getting bottlenecked by my slightly unusual set up.
I sadly want it all, Mac hardware and Gaming PC performance, we're getting closer, just wish Apple would make things a little bit easier for us (Nvidia/Bootcamp support).
@eightarmedpet I used rEFInd on the 2018 Mac mini yesterday after fully removing all drivers with DDU. rEFInd is more predictable in terms of eGPU detection (ADM eGPU connected to TB3 port closest to HDMI port). One thing I noticed is that H2D bandwidth through rEFInd is 2,1xx MB/s
vs 2,5xx MB/s when using automate-eGPU EFI due to Windows 10 1809. This may explain why you were able to get the 5K UltraFine going with automate-eGPU EFI but not rEFInd.
@itsage cheers for the extra info and testing! I had a hunch that would be the case, guess I’m stuck between a rock and a hard place due to the Ultrafine.
Think the mini was a bit of an impulse buy anyway, trying to cut down on computers not buy more! For now at least... thanks again champ!
@eightarmedpet I did more testing with Thunderbolt 3 monitor output through the UHD Graphics 630 iGPU. When I booted into Windows without the eGPU I get Thunderbolt 3 output on the HP Z27q monitor (via Thunderbolt 3 to dual DisplayPort adapter). When I booted into Windows with the eGPU, I do not get Thunderbolt 3 output with the same iGPU arrangement. So the eGPU does something to the amount of bandwidth available to the 5K Thunderbolt 3 connection. Below are AIDA64 screen captures on the same hardware (mini-eGPU prototype + R9 Nano). You can see the higher H2D speed through Thunderbolt 3 connection
using @goalque's automate-eGPU EFI due to Windows 10 1803.
|rEFInd Boot Manager||Automate-eGPU EFI|
So the eGPU does something to the amount of bandwidth available to the 5K Thunderbolt 3 connection. Below are AIDA64 screen captures on the same hardware (mini-eGPU prototype + R9 Nano). You can see the higher H2D speed through Thunderbolt 3 connection using @goalque's automate-eGPU EFI.
Interesting statements. As you may recall, I was able to get a Vega 64 & GTX 1070 to run simultaneously on Bootcamp Windows 10 (one eGPU per side - it just worked without any hacks). I didn't test AIDA bandwidth, but perhaps the results could have been interesting.
Odd to see a bandwidth difference between reFIND vs. automate-eGPU EFI. Are these results consistent and reproducible across multiple boots?
purge-wrangler.sh ✧ purge-nvda.sh ✧ set-eGPU.sh ✧ automate-eGPU EFI Installer
Troubleshooting eGPUs on macOS
Command Line Swiss Knife
eGPU Hardware Chart
Multiple Build Guides
Current: MacBook Pro RP560X + 480/R9 Fury/Vega 64 | GTX 780/1070
Previous: 2014 MacBook Pro 750M + 480/R9 Fury | GTX 780/980 Ti/1070
@mac_editor Thank you for questioning. I went through all recent builds and believe the lower H2D speed is actually due to Windows 1809 rather than rEFInd vs automate-eGPU EFI. Here are Mac mini builds with Breakaway Puck 560, and RX 590 + Node Pro which posted higher H2D speed. There are also examples from the 2018 MacBook Pro 13" with RX Vega 56 + Mantiz Venus, WX 9100 + Node Pro, and R9 Nano + Gaming Box. These builds were all on Windows 10 1803 and a mix of rEFInd and automate-eGPU EFI. I will upgrade the 2018 MacBook Pro 13" to Windows 10 1809 soon to confirm.
Update: I switched back to using automate-eGPU EFI on this Mac mini (W10 1809). The H2D speed did not improve. What changed was Thunderbolt 3 monitor output from the UHD graphics iGPU to the HP Z27q. The eGPU connected to TB3 port next to HDMI while the TB3 to dual DisplayPort adapter connected to TB3 port next to Ethernet port.
What changed was Thunderbolt 3 monitor output from the UHD graphics iGPU to the HP Z27q. The eGPU connected to TB3 port next to HDMI while the TB3 to dual DisplayPort adapter connected to TB3 port next to Ethernet port.
So you get better H2D to the eGPU when the Thunderbolt display is connected to the other Thunderbolt controller?
This means that DisplayPort output on a single Thunderbolt controller to a single port affects both ports of the Thunderbolt controller?
We've already seen that two ports are limited for PCIe traffic - 22 Gbps for each port but nowhere near 30 Gbps when you try to raid both ports together. I didn't think PCIe traffic on one port would be affected by DisplayPort traffic on the other port.
Thunderbolt controllers are not as fast as they could be, even if you allow for the 22 Gbps cap for PCIe traffic.