Apple doubles down on on-device AI performance with M5 chip: What's new

Apple said its M5 chip delivers up to 3.5x faster AI performance, 45 per cent better graphics, and higher efficiency across newly launched MacBook Pro, iPad Pro, and Vision Pro devices

Apple has introduced the M5 chip, marking a major upgrade in speed and efficiency. The new processor powers the newly launched MacBook Pro, iPad Pro, and second-generation Vision Pro headset. Apple said the M5 chip delivers up to 3.5x faster on-device AI performance compared with the M4, along with improved graphics, better power efficiency, and higher memory bandwidth for handling demanding tasks such as running large language models (LLMs) directly on the device.

Apple M5 chip: Details

Architecture

 

According to Apple, the M5 houses the world’s fastest CPU core, comprising up to 10 cores – four performance and six efficiency cores. It also features third-generation ray tracing and enhanced shader cores, making gaming and 3D rendering faster and more realistic.

 

The chip introduces an improved 16-core Neural Engine, a media engine, and higher unified memory bandwidth of 153GB/s, which is likely to allow smoother multitasking and advanced AI operations. The new GPU architecture in the M5 is designed to handle both AI and graphics-intensive tasks, said Apple.

 

The dynamic caching system ensures smoother performance in visually demanding apps and games. With Neural Accelerators built into each GPU core, the M5 is said to deliver a major boost to AI workloads.

 

Apple said the M5 combines faster graphics, the fastest CPU core, and higher unified memory bandwidth to bring improved performance across the MacBook Pro, iPad Pro, and Vision Pro. 

AI and developer features

 

With its next-generation GPU architecture, Apple said the M5 chip is designed with every compute block optimised for AI processing. The 10-core GPU includes a dedicated Neural Accelerator in each core, enabling faster AI performance than the M1.

 

Apple has also introduced Tensor APIs in Metal 4, allowing developers to directly program the Neural Accelerators for custom AI-driven applications. The unified memory architecture supports up to 153GB/s bandwidth, enabling devices to run large AI models fully on-device—an advantage over traditional chips that depend heavily on cloud-based computation.

 

With up to 32GB of unified memory, users can run demanding creative software such as Final Cut Pro and Adobe Photoshop side by side without lag. Apple said this level of efficiency puts the M5 far ahead of both previous M-series chips and conventional desktop processors. 

 

How M5 compares to previous models

• The M5 chip delivers up to 15 per cent faster multithreaded performance than the M4, driven by enhanced shader cores and third-generation ray tracing for more realistic graphics and smoother 3D rendering.

• With an upgraded 16-core Neural Engine and Neural Accelerators in every GPU core, the M5 achieves over six times faster AI performance compared with the M1, improving on-device machine learning efficiency.

• The M5’s 10-core GPU delivers up to four times higher peak performance and 45 per cent better graphics than the M4, offering faster rendering and richer visuals for gaming and creative workflows.

• Offering 30 per cent higher unified memory bandwidth at 153GB/s, the M5 ensures smoother multitasking and faster data handling, supported by an optimised media engine for better performance and efficiency.