Flaky Goodness

The Best Thing About eGPUs

May 27, 2020

A couple of years ago I was considering switching from a powerful desktop/small laptop setup to laptop-only. The increasing cost of CTO Mac hardware was making me question the need to maintain and upgrade two developer-spec Macs at the same time. And I was tired of the little wrinkles in syncing configuration and data despite having automated quite a bit of it. There really is something to be said for having everything on a single machine that you can pick up and take with you.

I was hesitant to lose the performance and stability of a desktop, and dreaded the hassle of plugging and unplugging cables and dongles every time I took my machine on the road. I was also skeptical that I’d be able to get a multi-monitor retina setup working with sufficient performance given the bandwidth limitations of Thunderbolt 3 and the hot/loud/relatively slow GPUs that are built into MacBooks.

Then a friend showed me his MacBook Pro setup with the Mantiz Venus eGPU chassis1 and I was all-in. I’ve been running this way since 2018 and it’s amazing.


Figure 1: Mantiz MZ-02 (Venus)

Inside the eGPU chassis I use an AMD Radeon Vega 562 that can handily output to both a Dell 5K display and a second Dell 4K display3 at the same time, over DisplayPort, without breaking a sweat3.

The eGPU chassis connects to the 15” MacBook Pro using a single Thunderbolt 3 cable. And it also provides 87W of USB-C Power Delivery for charging the laptop over the same cable.

And it also provides five (!) USB3 ports, all signaling over the same cable.

And it also provides gigabit Ethernet, also over that same cable.

It gets crazier. The Mantiz chassis also has an internal bay for a 2.5” SATA drive, which communicates (say it with me) over the same cable.

Why do they even call them eGPUs? They’re actually “single-cable docking stations that happen to have a PCIe slot where you can plug in a discrete GPU if you’re so inclined.” Or SCDSTHTHAPSWYCPIADGIYSI for short.

The hot-swappability works reasonably well and continues to improve with each macOS release. But the great undersold secret of eGPUs is the no-hassle I/O capability they can provide over a single Thunderbolt 3 cable4



As of this writing the Mantiz Venus is sold out and different eGPUs provide different I/O options. Some provide none at all other than power for charging. Do your research.


I was warned off the Vega 64 which is a little more power hungry and runs hotter.


I find Dell displays to be a lot nicer than the LG displays Apple promotes. I have a UP2715K (5K) which is no longer available, unfortunately, and a P2415Q (4K) which is even sharper with richer colors than the 5K.


I’m not sure about this but I don’t think that a single Thunderbolt 3 port has enough bandwidth to support both a directly-connected 5K and 4K display at the same time. With an eGPU setup though, the bandwidth requirement between the computer and the eGPU is lower and within the threshold. The computer is sending the eGPU instructions on what to draw. Most of the bandwidth appetite is between the eGPU and the monitors, where millions of rendered pixels are pushed 60 times a second. Thunderbolt 3 bandwidth allocation is a complex topic and different eGPU vendors have sliced up the bandwidth differently, with varying results. Intel’s Technology Brief goes into the intricacies but suffice to say that not all eGPUs perform well with all setups, and even software support can play a factor. I’m told that Final Cut Pro can actually perform worse with an eGPU setup than with built-in graphics in some configurations, for example.