With macOS 10.15 Catalina set to release later this month, I'm testing different eGPU configurations. Multi-eGPU is one that some Mac mini users would be interested in. The good news is you can load all four Thunderbolt 3 ports with an eGPU each.
System specs:
Late 2018 Mac mini – i5-8500B/UHD Graphics 630 iGPU/16GB RAM/256GB SSD
eGPU hardware:
NetStor HL23T-Plus + Radeon VII
Gigabyte Gaming Box + RX Vega Nano
VisionTek mini eGFX + R9 Nano
AKiTiO Node + GTX 780
Hardware pictures:
Installation steps:
While Radeon VII and RX Vega Nano work plug and play in macOS Catalina, GTX 780 and R9 Nano require workarounds. The latest version of Purge-Wrangler works great in 10.15 Beta to enable TI82 enclosures, older Radeon cards, and Nvidia cards which have native macOS drivers. Once installed these older cards work the same way as natively supported cards.
![]() |
![]() |
![]() |
Benchmarks:
I ran Luxmark 3.1 through each eGPU then all four together to see how they scale. The Radeon VII is the loudest of the bunch. Its OpenCL performance is also the strongest by a long shot.
GTX 780 | R9 Nano | RX Vega Nano | Radeon VII | All 4 eGPUs |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
Comments:
Apple must have realized many users cannot afford the upcoming Mac Pro. Therefore they gave the 2018 Mac mini Boot Screen support through the eGPU since dev beta 5. This is a much welcome improvement for a headless Mac. You can now switch between macOS and Boot Camp with ease.
• external graphics card builds
• best laptops for external GPU
• eGPU enclosure buyer's guide
2021 15" ASUS TUF Dash F15 [11th,4C,H] + RTX 3080 @ 32Gbps-TB4 (AORUS Gaming Box) + Win10 2004 // my 3rd RTX 3080 build [build link]
Yep, the 2018 Mac Mini is stealthy in a pretty great way. I love seeing the essential proof of concept here. One Mini can utilize a couple Radeon VII eGPUs (and even potentially four of them!). With graphics performance scaling quite usefully.
Adding a powerful dual-eGPU set up to the Mini is a nice looking avenue. Creating the equivalent of a re-born $3,299 “Mac Pro” that way is appealing. Especially with Metal-specific multi-GPU optimized renderers coming to Catalina. Accelerated Octane and Redshift GPU-rendering could be within reach of us individual Mac users.
Next test: 8 GPUs using Thunderbolt chaining...
Mac mini (2018), Mac Pro (Early 2008), MacBook Pro (Retina, 15-inch, Mid 2015), GA-Z170X-Gaming 7, Sapphire Pulse Radeon RX 580 8GB GDDR5, Sonnet Echo Express III-D, Trebleet Thunderbolt 3 to NVMe M.2 case
@joevt I will try daisy-chain a couple more eGPUs to AKiTiO Node Pro and Node Duo soon. Let's hope the entire setup won't trip the circuit breaker. 😀
• external graphics card builds
• best laptops for external GPU
• eGPU enclosure buyer's guide
2021 15" ASUS TUF Dash F15 [11th,4C,H] + RTX 3080 @ 32Gbps-TB4 (AORUS Gaming Box) + Win10 2004 // my 3rd RTX 3080 build [build link]
@itsage the eGPU support at boot applies only to Catalina? also, does it work with file vault enabled macs?
2018 Mac Mini Core i7 8700B 32GB RAM 512GB SSD + Razer Core X + AMD Radeon VII
@juan_c_yunis It's only for the 2018 Mac mini firmware (since Catalina dev beta 5). I don't use FileVault so I can't confirm but my guess is it should work. Make sure to use DisplayPort cable to the monitor. I've seen reports and experience HDMI cable/connection causing issues.
• external graphics card builds
• best laptops for external GPU
• eGPU enclosure buyer's guide
2021 15" ASUS TUF Dash F15 [11th,4C,H] + RTX 3080 @ 32Gbps-TB4 (AORUS Gaming Box) + Win10 2004 // my 3rd RTX 3080 build [build link]
Next test: 8 GPUs using Thunderbolt chaining...
I tried 6x eGPUs today. Thanks to @Mac_editor's Purge-Wrangler script, I could use Razer Core (TI82) and R9 Fury cards. The mix were 1x R9 Fury, 1x R9 Nano, 1x RX 580, 1x Vega Frontier, 1x Vega Nano, and 1x Radeon VII. They were all recognized in macOS. Unfortunately only the first 4x could be used. The 5th and 6th were not found by any applications. System Information was confused and couldn't load up. If I had it opened prior to connecting more than 4, it would stay open. This seems like a software limitation.
• external graphics card builds
• best laptops for external GPU
• eGPU enclosure buyer's guide
2021 15" ASUS TUF Dash F15 [11th,4C,H] + RTX 3080 @ 32Gbps-TB4 (AORUS Gaming Box) + Win10 2004 // my 3rd RTX 3080 build [build link]
Do 4 GPUs work when daisy-chaining instead of using all TB ports?
Edit: noticed in your image that one of the daisy-chained GPUs (Port 1, GPU 2) was working as expected, so that’s not the issue. It’s software limits as you surmise. They should increase this, and hopefully they do by the time Mac Pro lands.
purge-wrangler ✧ tbt-flash ✧ purge-nvda ✧ set-eGPU
Insights Into macOS Video Editing Performance
Master Threads:
2014 15-inch MacBook Pro 750M
2018 15-inch MacBook Pro
2019 13" MacBook Pro [8th,4C,U] + RX Vega 64 @ 32Gbps-TB3 (Mantiz Venus) + macOS 10.14.6 & Win10 [build link]
@mac_editor Port 1 and Port 3 were hosting two eGPUs through daisy-chaining. The R9 Nano was the 5th one recognized and Vega Frontier was the 6th one.
• external graphics card builds
• best laptops for external GPU
• eGPU enclosure buyer's guide
2021 15" ASUS TUF Dash F15 [11th,4C,H] + RTX 3080 @ 32Gbps-TB4 (AORUS Gaming Box) + Win10 2004 // my 3rd RTX 3080 build [build link]
@itsage So chained GPUs do work, but you can't have more than four total. How long can a chain be? I think you have enough to make a chain of 3.
Does the system_profiler command also fail? use "system_profiler -listDataTypes" to get a list of data types that can be output. Then try each one separately like this: "system_profiler SPThunderboltDataType" and "system_profiler SPDisplaysDataType".
There was a thread about max GPUs in macOS:
https://egpu.io/forums/pro-applications/macos-max-number-of-gpus/
Mac mini (2018), Mac Pro (Early 2008), MacBook Pro (Retina, 15-inch, Mid 2015), GA-Z170X-Gaming 7, Sapphire Pulse Radeon RX 580 8GB GDDR5, Sonnet Echo Express III-D, Trebleet Thunderbolt 3 to NVMe M.2 case