Apple unleashes a swarm of Mac Minions to share GPU resources for large AI tasks
Apple researchers have described Minions, a system that pools idle GPUs across Macs to run machine learning inference tasks for company teams. It matters because Apple says the system can ease delays caused by limited central GPU capacity, allowing some jobs to start within minutes instead of waiting for hours.
Minions divides work among Macs on the network, tracks tasks in PostgreSQL and uses a watchdog to detect failures and reassign work; results are uploaded to S3 and monitored through Grafana dashboards. Apple says some jobs previously faced queues of more than ten hours. The system focuses on making available hardware easier to use, while newer Mac mini and Mac Studio models can also combine compute and memory over Thunderbolt 5 for a single workload.
- Apple’s Minions shares idle Mac GPUs for AI inference.
- Some queued tasks can now start within minutes.
- A watchdog detects failures and reassigns work.
New here? Start with this
Graphics processing units, or GPUs, are chips that handle many calculations at once and are widely used to run artificial intelligence models. Large workloads can need substantial GPU capacity, which is often concentrated in shared data centres; when demand is high, teams may have to wait for access.
Apple’s Minions system is designed to spread parts of a job across Macs on a company network when their GPUs are idle. It coordinates the work and can reassign tasks if a machine fails. The idea matters because it offers another way to make computing capacity available for AI work, alongside using larger, centrally managed machines.