Recent reports indicate that Apple's upcoming A18 series processors will feature significantly enhanced Neural Processing Unit (NPU) performance, potentially surpassing the NPU capabilities of the current M4 processors. The A18 series is expected to debut with the iPhone 16 series this fall and will likely be produced using TSMC's advanced second-generation 3nm process (N3E).
During the Apple Worldwide Developers Conference (WWDC) on June 10th, Apple unveiled a groundbreaking AI system named Apple Intelligence. This innovative system will be integrated into the new operating systems for iPhone, iPad, and Mac, leveraging the enhanced capabilities of Apple's chips to understand and generate language and images, as well as unlock AI functionalities across various applications. To support these sophisticated AI features, the A18 series processors will require a robust NPU.
Previous iterations, such as the A17 Pro processor, featured a 16-core NPU with a performance benchmark of 35 TOPS (Tera Operations Per Second). However, with the increasing need for mobile devices to support large-scale models with billions of parameters, higher NPU performance is essential. Microsoft's definition of an "AI PC" includes an NPU with a minimum of 40 TOPS, suggesting that the A18 series will need to achieve even greater performance levels.
In May, Apple introduced the M4 processor in the new 11-inch and 13-inch iPad Pro models. The M4, also based on TSMC's second-generation 3nm process, includes 28 billion transistors (3 billion more than the M3), 4 performance cores, 6 efficiency cores (2 more than the M3), and a 10-core GPU. The NPU performance of the M4 has been notably increased from 18 TOPS to 38 TOPS, incorporating dynamic caching, hardware-level ray tracing, hardware-accelerated mesh shading, AV1 decoding, and AI acceleration.
Given these advancements, it is anticipated that the NPU performance in the A18 series processors will exceed that of the M4, potentially reaching up to 45 TOPS.