Apple has signed a driver for AMD or Nvidia eGPUs connected to Apple Silicon but there are some big caveats, and it won’t improve your graphics. Here’s what they’re for.
When Apple announced the use of eGPUs with AMD Radeon cards in 2016, we were pretty excited. Full support shipped in early 2017 and for a few short years, Thunderbolt provided an excellent graphics-accelerating one-cable dock to our MacBook Pros.
But even then, Apple has stubbornly prevented modern Nvidia GPUs from working with Macs. And, with the change to Apple Silicon, Apple effectively killed off any real use of an externally usable Nvidia GPU with its Mac lineup.
While the long-time feud between the two companies is still going on, Apple has made one small change that will be hugely beneficial to AI researchers. A third-party eGPU driver for AI research has been signed by Apple, making a setup far easier.
But, there’s a big limit. Video to an external monitor is not accelerated at all.
In an X posting on April 1, Tiny Corp declared that Apple had approved its driver for AMD and Nvidia cards in eGPUS for use on Apple Silicon. Using a Thunderbolt or USB 4-supporting eGPU enclosure, the setup now finally does something on a Mac again.
The instructions for the driver explain that it requires a Mac running macOS 12.1 or later, a spare USB4 or Thunderbolt 3 port or newer, and a supported GPU with an enclosure that delivers sufficient power to drive the card.
Supported cards are AMD RDNA3+ or Nvidia Ampere+ card at a minimum. Be mindful of power requirements, though.
After installing the card and initiating a driver installation, TinyGPU.app is installed, followed by the driver extension.
Users are then prompted to toggle the driver extension on in System Settings.
After that, the instructions differ based on the cards in use. While there’s one compiler setup that’s quite straightforward for AMD, Nvidia cards requires the use of Docker Desktop.
After that, a command can then be used to install a model that connects and is processed by the GPU.
An extremely specific use case
A new Nvidia-supporting eGPU driver for Apple Silicon is an interesting, considering the two companies aren’t exactly working together and haven’t for years. However, the driver is created by Tiny Corp, a different company.
The point of the driver is to enable GPU processing for AI uses. While Apple does have a good GPU and the Neural Engine for machine learning tasks, there are other ways to process them.
In this case, the driver allows users to employ external GPUs, which can be much more powerful or efficient than Apple’s GPU. The upshot is that AI researchers have a way to run models even faster, or to improve the way that large language models are created in the first place.
By approving the driver, Apple also makes it safer for anyone experimenting with AI on eGPUs. The approval means that developers can use the driver without needing to disable Apple’s System Integrity Protection (SIP) beforehand.
Since previous work in this area required SIP to be disabled, this change will make things much more secure for researchers working on their Macs.
Apple’s approval of the driver enables eGPU use in a very limited fashion, and with multiple hoops to jump through. It’s a far cry from Apple signing Nvidia’s own drivers, or using the enclosure to accelerate graphics on any machine.
However, it is a big step forward for AI researchers.