Apple’s latest claims highlight that iPhones can run large language models (LLMs) with 14 billion active parameters, while Macs can support AI models with over 1.6 trillion parameters under specific conditions.
This capability is made possible by Apple’s custom silicon, which integrates dedicated neural processing units (NPUs). For iPhones, the A17 Bionic chip features a 16-core Neural Engine that can handle up to 17 trillion operations per second (TOPS). This allows iPhones to run LLMs locally, reducing latency and improving privacy. For Macs, the M1 Ultra chip, with its 32-core GPU and 16-core Neural Engine, can support much larger models, making it suitable for heavy AI workloads.
For mini PC and local AI users, this means that Apple’s hardware is now capable of running significant AI tasks locally, which is useful for privacy-conscious users or those with limited internet access. However, the specific conditions under which these capabilities are achieved are not fully detailed, so it’s important to verify these claims with real-world testing.
