Apple launched its M3 Ultra chip last week to some surprise, positioning it as the best performer for developers and professional users who want on-device AI inferencing tasks that are expected to be more common in the years ahead.
Despite having an M4 Max chip out last year, the company touts the upgraded M3 Ultra as its most powerful chip to date. It sports a high-performance graphics processing unit (GPU) optimised for the parallel computations inherent in AI inferencing.
At the core of the M3 Ultra lies Apple’s enhanced neural engine, a key driver for accelerated machine learning tasks. This, combined with the chip’s unified memory architecture, promises to eliminate critical memory bottlenecks – a common challenge in AI workloads involving massive datasets.
The M3 Ultra uses Apple’s UltraFusion packaging architecture, which connects two M3 Max dies via over 10,000 high-bandwidth, low-latency connections. This design allows the system to treat the combined dies as a single, unified chip to deliver improved performance.
Featured in the new Apple Mac Studio, the M3 Ultra enables users to run large language models (LLMs) with over 600 billion parameters directly on device. This capability was previously confined to cloud-based solutions.
So who would be the users? They are likely to be AI developers and other power users.
Developers gain a powerful platform for creating and deploying sophisticated AI applications directly on Apple devices, while users benefit from faster, more responsive AI-powered tools, boosting productivity and creative workflows.
The new chip’s ability to utilise very large amounts of unified memory allows for the running of very large AI models locally. This is a key differentiator for professionals who require processing of very large data sets.
Research firm IDC anticipates a growth in AI PCs, particularly driven by commercial upgrades. It expects a more robust ecosystem for GenAI-capable PCs to emerge by late 2025, leading to a majority of new PCs being GenAI-capable by 2026.
Apple is not alone in launching AI-powered processors and computers. It will face competition from high-performance processors like Intel’s “Arrow Lake-H” Core Ultra, AMD’s Ryzen 9 7945HX, and Qualcomm’s Snapdragon X Elite.
Notably, Apple has not announced plans to introduce the M3 Ultra chip in the MacBook Pro line. The M3 Ultra is designed for users with extreme performance needs such as video editing and AI development. Its high power consumption and thermal output also does not make it suitable for laptops like the MacBook Pro.
For now, at least, Apple is reserving the Ultra chips for desktop systems that can accommodate the higher power and cooling requirements. For consumers with lower high performance needs, expect to see Apple MacBook Pros powered by the M4 Max chip, which was released late last year.
