TinyCorp Runs Nvidia GPU Off Apple Silicon Mac via USB4 dock (www.tomshardware.com)

TinyCorp, an AI startup, has ported its eGPU driver work to USB4/Thunderbolt 4 and released drivers that let Apple M‑series MacBooks run Nvidia RTX 30/40/50 GPUs via external docks (demoed with an ADT‑UT3G). The drivers are targeted at AI development—not display output—so you can run compute workloads (Tinygrad shown running on an attached RTX) but not use the card for the MacBook’s screen. TinyCorp shared install hints on X: disable SIP, install their driver (extra/usbgpu/tbgpu), install the NVK compiler via brew install tinymesa, and run their test harness (e.g., DEBUG=2 NV_NAK=1 NV=1 python3 test/test_tiny.py TestTiny.test_plus). Support requires Nvidia GPUs with a GPU system processor (no GTX series); RTX 20‑series may be possible with extra work. TinyCorp previously enabled AMD eGPUs on Apple silicon over USB3, and RDNA2/3/4 AMD cards reportedly work too. Why it matters: USB4/Thunderbolt 4’s native PCIe and higher bandwidth make practical external GPU compute on Apple silicon, so developers can tap high‑end RTX cards (even future RTX 5090‑class accelerators) to run local LLMs and other ML workloads far faster than the onboard iGPU. Caveats are significant: this is experimental, requires low‑level system changes, limited hardware compatibility, no display support, and potential thermal/power/driver stability and legal risks—so it’s primarily useful to developers who need local GPU compute on MacBooks rather than general consumers.