Mac minis can now use eGPUs

An external graphics processor is a graphics card in a separate box, linked to a computer over Thunderbolt or Universal Serial Bus 4. Apple silicon Mac minis can now use one for artificial intelligence compute through TinyGPU, a driver Apple approved for macOS driver extensions. (docs.tinygrad.org, xda-developers.com) TinyGPU’s setup page says it works on macOS 12.1 or later over Thunderbolt or Universal Serial Bus 4, and supports Advanced Micro Devices RDNA3-and-newer cards and Nvidia Ampere-and-newer cards. The install flow uses a system prompt in macOS Settings to enable the driver extension, instead of asking users to bypass Apple’s normal driver controls. (docs.tinygrad.org) That changes what a Mac mini can do for local inference, which is the step where a model answers prompts after it has already been trained. XDA reported on April 11 that OpenClaw users have been leaning on Mac minis as small always-on hosts for local agents, and that the new driver lets those systems attach larger graphics cards in external docks. (xda-developers.com) The timing lines up with a broader run on higher-memory Mac desktops. On April 11, 9to5Mac reported that Apple’s online store had marked the Mac mini with 32 gigabytes of memory as “currently unavailable,” along with multiple Mac Studio configurations. (9to5mac.com) MacRumors reported the same day that Mac mini configurations with 32 gigabytes or 64 gigabytes of memory, and Mac Studio configurations with 128 gigabytes or 256 gigabytes, could no longer be ordered in the United States. Other configurations were still listed, but with delivery estimates ranging from one to three months. (macrumors.com) The shortage appears concentrated in memory-heavy builds, which are also the versions people buy when they want to run bigger models locally. MacRumors said the pattern points more clearly to the global memory-chip squeeze tied to artificial intelligence server demand, while 9to5Mac said an M5 refresh remains a possibility. (macrumors.com, 9to5mac.com) There are still limits. TinyGPU’s own documentation lists only newer Advanced Micro Devices and Nvidia architectures as supported, and recent GitHub issue reports show some individual cards still failing to initialize on Apple silicon Macs even when the driver extension is active. (docs.tinygrad.org, github.com, github.com) So the Mac mini story now has two tracks at once: more ways to add outside graphics power, and fewer top-end memory configurations available to buy. For people building small on-device artificial intelligence boxes, the bottleneck has shifted from “can this Mac use an external graphics processor at all” to “which Mac and which card can you actually get.” (docs.tinygrad.org, macrumors.com, 9to5mac.com)