IT Home On October 16, a tech media site AppleInsider posted a blog post yesterday (October 15) stating that Apple’s research team unveiled a system study called Minions. They attempted to combine idle Mac GPUs into a dynamically scheduled “virtual GPU cloud” to run large-scale machine learning inference tasks.
According to a blog post cited by IT Home, traditional AI inference usually relies on centralized GPU clusters, which face issues such as high costs, slow expansion, and long waiting times; whereas on Macs, GPUs are often left idle for long periods during daily office work, browsing, and creative activities.
The Apple company explained the Minions solution in its paper, hoping to reorganize this “invisible computing power” so that multiple Macs can collaborate to complete AI tasks without purchasing additional GPUs.
According to the paper overview, the system consists of four key components:
Minion Server is responsible for receiving tasks, breaking them into sub-tasks, and coordinating scheduling;
PostgreSQL metadata storage records tasks and device status;
Minion Executor running on each macOS device executes inference locally through the Apple Silicon GPU;
Minion Watchdog monitors situations such as device disconnection, sleep, or power outage, and reassigns unfinished tasks to other available devices.
In terms of technical implementation, Minions adopts a “pull” task model: devices actively request executable sub-tasks from the server, and an atomic mechanism is used to prevent multiple devices from processing the same task simultaneously. The inference results can be written into object storage, and the system provides a real-time monitoring panel based on Grafana, which helps the team understand the number of online, busy, and idle devices.
The testing results disclosed by Apple show that some reasoning tasks that previously had to wait over 10 hours due to insufficient centralized GPU resources can now be executed within a few minutes under Minions; the team does not need to add new GPU hardware, and infrastructure costs are also reduced.
However, this technology currently seems more like a research and engineering solution for corporate internal use, rather than a formal product for ordinary consumers. The research team also pointed out that when using others' equipment across different departments, data confidentiality and governance are important concerns.
Reference
Advertising statement: The text contains external link pointers (including but not limited to hyperlinks, QR codes, passwords, etc.) intended to provide more information, save selection time, and the results are for reference only. Articles on IT Home that include external links all contain this statement.