New Mac Mini plus Nvidia eGPU viable Workstation?


Are you getting linear scaling with daisy chaining? same as if they were internal?

When will manufacturers produce a multi gpu external box? They all seem to have only one card or slot.


I do get linear scaling, even better performance than having them internal.
There is one double egpu box by akitio, but it’s for single slot (less powerfull ) gpus at the moment. At some point I imagine we will see multi gpu enclosures, (that would be awesome) even better if they support water cooling ! Maybe with thunderbolt 4? Time will show…

Konstantinos D


The eGPU I bought, the Razer Core X, has only one thunderbolt port. So I guess I would need to have that as the last link in the daisy chain, and use another eGPU with two ports to start the chain?


Yes, I bought the akitio node pro which has 2 tb3’s and double the bandwidth so you could daizy chain one more egpu

Konstantinos D


What is your OSX, NV driver and CUDA version? I have a 2016 MBP 15” w TouchBar on 10.13.6 and I can’t get my system to recognise more than 2 GTX cards attached, whether in a multi-slot enclosure or in separate enclosures(x1 1080 and x2 780). I have a Mantiz Venus and a 8-Slot CUBE3 from MaxExpansion, both with only one TB port so I am unable to daisy chain.

I would love to know which mod you have applied from


I’ve used this one script but in my case I had to put the --beta flag and install manually because my mac was too new and not supported yet. It all worked fine, now I have 3 Titan x’s egpus and I’m expecting 3 1080 ti’s more.
Also, forgot to mention, I have the 2018 MacBook Pro 15” with touchbar 32gb ram, 4 x thunderbolt 3 ports, and High Sierra 10.13.6

Konstantinos D


Thanks I’ll give that script another try!

Please update once you have your x6 GTX running! You are daisy chaining 6 enclosures to one TB port?


No, I’m doing 6 devices on 3 tb3 ports, so it’s two for each port. I could probably do two more for a total of 8. Any more than that I would have to test, because the akitio node pro has 10 Gb of bandwidth (2 x tb3 ports) but the akitio node has 5 ( end of the chain with one tb3 port) I have 3 of each. Maybe we would have some bandwidth issues , maybe not. If not, then you could go crazy and chain 3-4 cards to each port for en extreme of 16!! I’ll have to do some tests when I get them with octanebench…

Konstantinos D


Tb3 have 40gb/s bandwidth(theoretical), not 10;)


Yes, but the akitio node Pro has 10. Unfortunately tb3 has that potential but it’s not necessarily being used by all the external devices on their end.

Konstantinos D


Sounds weird, where I can read about this limitation? even on their website says up to 40gbps and I’ve seen SSD performance exceeding 20gbps. Maybe there’s something I’m missing.


Here, check the io bandwidth Link
In the egpu list (5Gb or 10Gb). Some ssd ‘s are pretty fast, I know, but not all io connections are created equal, unfortunately.

Konstantinos D


I’ve seen the link and readed every note and related links, seems that numbers refers only to the USB3 I/O speed as far as I can tell(and that sounds correct). This post from the notes explain eGPU/USB bandwidth :slight_smile:


Oh I see… ! As in many times in life, I was wrong :slight_smile: it’s better you were right though…
Good to have the full 40 , that means I can daizy chain more cards.

Konstantinos D


Be aware that even if your eGPU is using full TB3 bandwidth you may experience limitation on the computer side, check if every port on your laptop support full speed(this depends by the PCIe lanes available from the CPU).
That being said, in general the bandwidth required by GPU renderer is very limited and should affects mostly the loading time of the scene, so you should be fine even using several cards.
BTW its very good to know that daisy chain works also with GPU, this open a great number of possibility. An eGPU enclosure cost money for sure but this way you can add far more cards compared to a traditional tower and still maintain a system quite and cool(when not using all of the GPU), and even portable like in your case.


Yes, exactly! So much opportunity/scalability etc with this modular system, it’s hard to go back to internal gpus… fortunately my MacBook Pro 2018 supports the full speed on the ports so I should be fine . I’ll run tests when I get all the cards too see how many egpus one tb port can handle before we start seeing degradation in performance. I also hope the new Mac pros ( whenever they arrive) have even more full speed tb ports.
Interesting times…

Konstantinos D


I hope the Mac Pro does too, and that nVidia drivers for the RTX series and Mojave support comes with it!


Ok I have an update!
I received the 3 akitio node pros and one 1080ti’s, waiting for more. I have good news and bad news.
The bad: I cannot mix 2 different architecture Gpu’s. No matter which configuration, it only recognizes the one you plug in first, in my case, the titan x’s ( maxwell ), or the 1080 ti ( pascal). This needs further investigation and maybe is solvable with a tweak, the gurus at should know.
The good: I can daizy chain the 3 titan x’s just fine on one port! This gives me a theoretical 3 egpu’s per port 12 in total, maybe even more who knows. I’ve got just 20 less points in octanebench by doing that , no big drop, 430 for the 3 titans instead of 450 could be just a marginal difference. I am now expecting 5 more 1080 ti’s for a octanebench of around 1300 with the 6x 1080 ti’s. Around 1750 if I find a way to connect the Titan x’s!!
So in a nutshell yes, I find that a viable workstation even with the Mac mini . I just hope that they release Nvidia drivers in the future for Mojave and they don’t drop support altogether otherwise I’ll have to stay in high Sierra forever… (the OS not the place :wink: )

Konstantinos D


Really exciting to read this. Thank you for sharing your experiences.


Once you’ve got it all running the way you want, can you make us a video/screen capture showing how Octane performs?