Apple's M1 is up to 3.6x as fast at training machine learning models

Apple's M1 chip, Apple's first custom-designed silicon for Mac, exhibits promising performance in machine learning tasks compared to its Intel-based predecessors, particularly in computer vision workloads. Using Apple's CreateML for testing due to incomplete TensorFlow support, the M1 outperformed the Intel Core i5 by 3.64 times in training a YOLOv2 object detection model, although it lagged behind the Intel Core i9 with a discrete Radeon Pro 5500M GPU. Despite the M1's hardware potential, software optimizations are still needed to fully leverage its capabilities, as the GPU utilization was limited during testing. As software improvements are expected, the M1 chip is poised to outperform traditional setups, making it a compelling option for future machine learning tasks once these enhancements are realized.