Apple Announces Mac-Focused Developments During AI Week
Apple concluded its AI-focused event week with announcements regarding local model execution capabilities on Mac devices and a partnership with Meta.
Apple has expanded its focus on artificial intelligence by announcing new tools and partnerships aimed at running AI models locally on Mac computers in a cost-effective and responsible manner.
Artificial Intelligence and Mac Integration
Apple researchers shared Apple Intelligence plans in China by developing a new tool that turns Mac devices into distributed inference data centers.
Working in partnership with Meta, the company enabled Muse Spark Open models to be run locally on Mac Studio systems.
MLX Updates and Hardware Support
The machine learning team introduced a new version of MLX, which allows large language models to run locally by taking advantage of Apple silicon's unified memory architecture.
MLX enhancements include software support for Neural Accelerators on M5 family GPU chips and low-latency communication via RDMA over Thunderbolt.
Supercomputer Performance
According to a report published by Apple, a cluster of four Mac Studios running Muse Spark Open can generate more than 36 tokens per second.
The Meta and Apple partnership allows businesses to run cutting-edge models locally while securing corporate data.
Testing Processes and Workflows
Meta noted that the M5 Ultra's integrated GPU can scale up to 512 GB per Mac Studio, allowing the machines to act as a single supercomputer with one terabyte of unified memory.
Conducted tests included agentic workflows such as bug fixes for a Mastodon client and the routing of the DaVinci Resolve application.