AI Cluster: Four Macs with 2 TB Can Run Kimi K2 Thinking Model
Apple has enhanced macOS to facilitate the creation of AI clusters from multiple Mac machines. The new macOS Tahoe version enables the utilization of combined hardware with low latency via Apple's MLX interface and Exo Labs software.
Apple has updated its macOS operating system to enable the formation of AI clusters by connecting multiple Mac computers. The latest macOS Tahoe version allows combined hardware to be utilized with low latency through Apple's proprietary MLX AI interface and Exo Labs software.
In a demonstration, Apple connected four Mac Studio M3 Ultra machines into a Thunderbolt 5 network. Each machine was equipped with 512 GB of RAM. This setup was capable of running the Kimi K2 Thinking AI model, developed by Chinese company Moonshot AI, which exceeds one terabyte in size. The model is considered to be nearly on par with GPT-5.
However, this solution comes at a significant cost. A single Mac Studio M3 Ultra with 512 GB of RAM and a 1 TB SSD is priced at approximately 11,674 euros. A cluster of four such machines would amount to nearly 47,000 euros, offering researchers and professional developers with substantial budgets the capability to run large local AI models.
The cluster's peak power consumption was recorded at 500 watts, which is a tenth of that for a comparable GPU cluster. Thunderbolt 5 facilitates an 80 Gb/s data transfer rate and low latency. This technology can also be applied to the more affordable Mac Mini M4 Pro, though it is limited to 64 GB of RAM per unit.