Apple researchers describe pooling idle Mac GPUs for AI inference

Apple researchers presented Minions, a distributed system that pools idle graphics processors in Apple Silicon Macs to run machine learning inference, the process of producing outputs from trained models. The paper describes a virtual GPU cloud that uses existing macOS devices rather than requiring additional GPU purchases. It supports flexible task execution and fault tolerance across the participating machines. The authors say their experience shows that idle Macs can provide a practical inference platform where centralized GPU resources are constrained, framing the system as a way to use hardware that otherwise remains largely idle.