A thousand patches and a long-awaited SoC

The Linux 7.2 kernel has officially integrated support for five new system-on-chip designs, headlined by the eagerly awaited Apple M3 port. The milestone comes nearly three years after the silicon's debut, thanks to painstaking work by Asahi Linux and other open-source contributors.

Apple M3 on Linux: current state

A stable boot on M3-based Macs is a major step but not a complete solution. The kernel recognizes the SoC and manages CPU, memory, and basic peripherals; GPU support is still under development, and the Neural Engine’s potential remains largely untapped without dedicated drivers. Today an M3 machine can run Linux with competitive CPU performance but without hardware acceleration for machine learning workloads. Asahi’s roadmap aims to close these gaps gradually, but it will take time.

Why it matters for local inference

For those deploying LLMs on-premise, Apple’s unified memory architecture holds growing appeal. Devices like Mac Studio with M2 Ultra (and potentially M3 Ultra) offer up to 192 GB of shared memory, enough to load large models without multi-GPU clusters. In a context of data sovereignty and full stack control, running self-hosted inference on Linux hardware is a compelling goal. The trade-off, however, is stark: without a fully supported GPU, M3 inference under Linux defaults to CPU, limiting throughput and latency. The situation will improve as drivers mature, but today anyone seeking a production-ready LLM serving node with Linux would hardly choose M3 over servers with NVIDIA or AMD GPUs, or ARM alternatives like Ampere. For experimentation and small-scale fine-tuning, though, the ecosystem could become attractive soon.

A shifting landscape

The addition of four other SoCs in kernel 7.2 signals a vibrant SoC side, propelled by the need for heterogeneous architectures for edge computing, robotics, and local infrastructure. In the world of on-prem LLMs, solutions to reduce single-vendor dependency are multiplying. Seeing M3 boot Linux, even in a still-partial fashion, broadens the horizon and sets the stage for Macs to become Linux workstations for AI developers, with all the benefits of open-source toolchains. For those tracking self-hosted deployments, progress is worth watching closely, because the line between consumer devices and on-prem infrastructure keeps blurring.