While the world of chips is mostly about how Nvidia, Broadcom, AMD or TSMC equip large data centers for OpenAI and Co, when it comes to AI chips, Apple is primarily concerned with the local: Apple has introduced its new M5 processor, which, according to the company, is intended to bring a significant leap in artificial intelligence performance. The chip is now being used in the 14-inch MacBook Pro, iPad Pro and Apple Vision Pro.
Technical innovations at a glance
Table of Contents
The M5, manufactured in the third generation of the 3-nanometer process, has a fundamentally revised GPU architecture. The most important innovation: Each of the ten GPU cores now contains a dedicated neural accelerator. According to Apple, this results in more than four times the peak GPU computing power for AI applications compared to its predecessor, the M4.
The CPU consists of up to ten cores – six efficiency and up to four performance cores. Apple speaks of up to 15 percent higher multithreading performance compared to the M4. The 16-core Neural Engine has also been revised and is intended to handle AI tasks more energy-efficiently.
Importance for local AI processing
The increased memory bandwidth is particularly relevant for processing AI models directly on the device: at 153 GB/s, the M5 offers almost 30 percent more than the M4. The unified memory architecture allows the CPU, GPU and Neural Engine to access a shared memory pool – with up to 32 GB capacity on the M5.
According to Apple, this configuration allows larger AI models to be run locally without having to rely on cloud services. This should significantly speed up applications such as running large language models via platforms such as webAI or diffusion models in apps such as Draw Things.
Software integration and developer support
Apple emphasizes integration with its own software frameworks such as Core ML, Metal Performance Shaders and Metal 4. Applications that use these APIs should automatically benefit from the higher performance. Developers can also program the Neural Accelerators directly via Tensor APIs in Metal 4.
Apple’s in-house AI system “Apple Intelligence” also benefits from the new chip. Functions such as Image Playground or the Writing Tools should be accelerated by the higher memory bandwidth and the faster Neural Engine.
Further improvements
In addition to AI performance, graphics capabilities have also been expanded. Apple claims up to 30 percent higher graphics performance compared to the M4. The third generation ray tracing engine is said to enable a performance increase of up to 45 percent in corresponding applications.
With the Apple Vision Pro, the M5 leads to a ten percent higher number of pixels on the micro-OLED displays and enables refresh rates of up to 120 Hz.
The new devices with M5 chip can now be pre-ordered. During the presentation, Apple also referred to the energy efficiency of the chip, which is part of the “Apple 2030” sustainability strategy, with which the company wants to achieve CO2 neutrality by 2030.
