Apple M3, M3 Pro and M3 Max: Chip Analysis

Apple's M3, M3 Pro and M3 Max, announced on 30 October 2023, were the first personal computer chips built on a 3-nanometre-class process, TSMC's first-generation N3 (widely referred to as N3B). The M3 Pro is the surprise of the family: it has fewer performance cores and less memory bandwidth than the M2 Pro it replaced. The video below is an in-depth analysis of the three chips, with a silicon deep-dive, die-shot analysis and a closer look at the CPU, GPU, NPU and the process node.

Key takeaways

• All three chips use TSMC's first 3nm-class process, N3B.
• Apple's published transistor counts are about 25 billion (M3), 37 billion (M3 Pro) and 92 billion (M3 Max).
• The M3 Pro moves to six performance and six efficiency cores and a narrower memory bus than the M2 Pro.
• The GPU adds Dynamic Caching, hardware-accelerated ray tracing and mesh shading.

Video chapters

• 0:00 Intro
• 0:47 M3 silicon analysis
• 4:17 M3 Pro silicon analysis
• 6:26 M3 Max silicon analysis
• 8:25 Why is the M3 Pro a downgrade?
• 10:47 NPU deep-dive
• 12:16 CPU deep-dive
• 13:17 GPU deep-dive
• 14:44 GPU architecture and the Apple family 9 GPU
• 16:22 TSMC N3B process node
• 18:44 Wrap-up

The M3 family at a glance

Chip	Transistors	CPU cores (max)	GPU cores (max)	Max unified memory	Memory bandwidth
M3	25 billion	8 (4 performance, 4 efficiency)	10	24 GB	100 GB/s
M3 Pro	37 billion	12 (6 performance, 6 efficiency)	18	36 GB	150 GB/s
M3 Max	92 billion	16 (12 performance, 4 efficiency)	40	128 GB	400 GB/s (300 GB/s on the lower bin)

Figures are as published by Apple at launch in October and November 2023, summarised here and not re-measured. All three chips have a 16-core Neural Engine.

Why the M3 Pro looks like a downgrade

The M2 Pro offered up to eight performance and four efficiency cores and 200 GB/s of memory bandwidth. The M3 Pro moves to six performance and six efficiency cores and 150 GB/s. Apple's position was that the new cores are faster, so overall performance still rises, while the smaller die and narrower memory interface cut cost and power. The trade-off is that the M3 Pro moves further away from the M3 Max, which keeps the wide memory bus and many more GPU cores. The video's die-shot section shows where that silicon area went.

What is new in the GPU

• Dynamic Caching: on-chip memory for the GPU is allocated at run time according to what each task actually uses, instead of reserving the worst case in advance. Apple says this raises average GPU utilisation.
• Hardware ray tracing: dedicated hardware for ray intersection, a first for Apple's Mac chips.
• Mesh shading: a more flexible geometry pipeline for complex scenes.

In Apple's Metal documentation this generation corresponds to the Apple9 GPU family, which the video covers from 14:44.

About the process node

TSMC's first 3nm process, usually called N3B, was used by Apple for both the M3 family and the A17 Pro in 2023. It was followed by N3E, which relaxes some design rules for better yield and cost. Because N3B and N3E are not design-rule compatible, chips moving from one to the other need re-implementation, which is one reason later Apple chips moved to N3E. The video's process section, from 16:22, discusses N3B in more detail.

FAQ

What process are the Apple M3 chips made on?

TSMC's first-generation 3nm-class process, widely reported as N3B.

How many transistors does the M3 Max have?

Apple quoted 92 billion transistors for the M3 Max at launch.

Is the M3 Pro slower than the M2 Pro?

Not in general. It has fewer performance cores and less memory bandwidth, but faster cores. Results depend on the workload, especially for memory-bandwidth-heavy tasks.