TinyGPU is presented as a way to use supported AMD and NVIDIA GPUs on macOS over USB4/Thunderbolt with tinygrad. The docs list macOS 12.1+, a USB4/Thunderbolt port, and supported GPUs of AMD RDNA3+ or NVIDIA Ampere+; they also describe separate setup paths for the driver, AMD compiler tooling, and NVIDIA compiler tooling.
In this video, I test one of the wildest Mac hacks I’ve seen in a long time: plugging an NVIDIA or AMD GPU into an Apple Silicon Mac and actually trying to make it work.
This is based on TinyGPU from tinygrad, which aims to let supported external GPUs run on macOS over USB4 / Thunderbolt. So this is not theory. This is a real attempt to push Apple Silicon beyond the limits Apple intended.
I walk through:
• what TinyGPU is
• what hardware is supported
• how the driver setup works
• AMD vs NVIDIA setup differences
• whether this is actually usable
• and what this means for the future of Apple Silicon power users
If this works reliably, it opens up a very different conversation around external compute, AI workloads, experimentation, and whether Apple’s GPU lock-in is as final as people think.
tinygrad TinyGPU docs:
https://docs.tinygrad.org/tinygpu/
#AppleSilicon #MacMini #eGPU #NVIDIA #AMD #RTX5090 #Tinygrad #TinyGPU #MacHack #AI #MachineLearning #GPU #Thunderbolt #USB4 #Tech