Apple Approves First Third-Party GPU Driver for Apple Silicon Macs

For the first time since Apple transitioned to its own silicon in late 2020, Mac users can now connect external Nvidia and AMD graphics cards to their machines through an officially sanctioned driver. Apple has approved TinyGPU, a third-party GPU driver extension developed by George Hotz's AI startup Tiny Corp, enabling Apple Silicon Macs to harness discrete GPU compute power over Thunderbolt and USB4 connections. The approval, announced on March 31 and widely covered in the days since, represents a landmark moment in Apple's historically closed hardware ecosystem.

TinyGPU is built on top of Tiny Corp's tinygrad neural network framework and ships as a standard macOS application that installs an Apple-approved DriverKit extension. Crucially, it requires no kernel extensions and no disabling of System Integrity Protection — the security mechanism that has long prevented unofficial hardware hacks on macOS. Users can enable the driver through a simple toggle in System Settings under General > Login Items & Extensions > Driver Extensions. The extension supports AMD RDNA3 and newer GPUs, including the Radeon RX 7900 XTX and RX 9000 series, as well as Nvidia Ampere architecture and newer cards across the RTX lineup.

Benchmark results published by Tiny Corp demonstrate practical real-world performance. A Mac mini equipped with Apple's M4 chip, connected to a Radeon RX 7900 XTX over Thunderbolt/USB4, achieved 18.5 tokens per second running Qwen 3.5 27B — a 27-billion-parameter large language model. While Thunderbolt bandwidth constraints mean the setup cannot match native PCIe throughput, the speeds are more than sufficient for interactive AI inference tasks. The current implementation is purpose-built for AI compute workloads rather than general graphics acceleration, meaning it does not yet support display output or consumer gaming scenarios.

The road to official approval was a long one. Tiny Corp first demonstrated what it called the "world's first" AMD GPU driven over USB3 from an Apple Silicon Mac in May 2025. By October of that year, the team had Nvidia RTX GPUs running on a MacBook Pro M3 Max over USB4. The progression from proof-of-concept to Apple-sanctioned DriverKit extension represents more than a year of engineering custom userspace GPU drivers — work that many in the developer community had assumed Apple would never permit. The driver requires macOS 12.1 Monterey or later, with AMD users needing the HIP compiler and Nvidia users requiring Docker Desktop for the NVCC compiler toolchain.

The strategic implications of Apple's decision may prove more consequential than the technical achievement itself. By approving a compute driver that explicitly supports Nvidia hardware — a company Apple severed GPU ties with years ago — Cupertino appears to be acknowledging the growing importance of GPU compute in the AI era. For researchers and developers who have invested in Nvidia or AMD hardware but prefer macOS as their primary operating system, TinyGPU eliminates the need to maintain separate Linux or Windows machines for local AI inference work.

The timing is significant. As AI models grow in capability and the demand for local inference rises — driven by privacy concerns, latency requirements, and the mounting costs of cloud compute — the ability to pair Apple Silicon's power-efficient architecture with the raw compute of discrete GPUs creates a compelling proposition. Researchers can now run large language models locally on their Macs without relying on cloud services, potentially reducing both costs and data exposure.

For Apple, the approval of TinyGPU could mark the beginning of a broader shift in platform strategy. The company has spent years building out its own GPU compute capabilities through Metal and the Apple Neural Engine, but the AI research community overwhelmingly relies on Nvidia's CUDA ecosystem. By allowing third-party GPU compute drivers through DriverKit's secure, sandboxed framework, Apple may be charting a middle path — maintaining its security standards while opening the door to the hardware diversity that AI researchers demand. Whether this remains a narrow exception for compute workloads or signals a wider reopening of the Mac platform to external GPUs remains to be seen, but for the millions of developers and researchers in the Apple ecosystem, according to TechRadar, the six-year eGPU drought is officially over.