Apple Unveils M6 and M5 Ultra Chips, Enhancing AI Performance and Memory

Cook’s final performance before retirement truly emptied the magazine and went all out!!

The first 2nm process chip, M6, has arrived.

However, the new M6 process is not the strongest; the strongest is the newly released M5 Ultra, which is the most powerful M-series chip in Apple's history.

Although it uses a 3nm process, Apple specifically highlights its groundbreaking four-die stacking technology—

It also represents an application of Huawei's Tao Law.

Moreover, these two chips were directly integrated into the new Mac—

The Mac mini with M6/M5 Pro starts at CNY 6,999 in China; the Mac Studio upgraded with M5 Max/M5 Ultra starts at CNY 19,999 in China.

The exterior is still the same familiar design, but inside, we're already making big strides in AI.

The AI performance of the M6 Mac mini has increased by up to 4 times.

The M5 Ultra Mac Studio features up to a 36-core CPU, 80-core GPU, and supports up to 512GB of unified memory.

More interestingly, as soon as this wave of AI computing power was released, the memory couldn't keep up...

Due to the global memory shortage, compounded by intense local AI demand for high-memory systems, the 512GB Mac Studio is now scheduled for delivery by the end of October.

Additionally, purchase limits have now been implemented.

This this this... Cook really used up all his ammunition before retiring from Apple.

Before leaving, I gave Mac another big boost!

Let’s start with M6.

The reason for highlighting this is that M6 truly brings Apple's chip technology into the 2nm era!! (Cheering.jpg)

The M6 is Apple's first chip manufactured using a 2nm process, with the CPU expanding from 10 cores in the previous generation to 12 cores.

Includes 2 Super Cores, 4 Performance Cores, and 6 Efficiency Cores; the GPU has also been upgraded to 12 cores, with unified memory bandwidth increased to 170 GB/s.

But a more noticeable change is reflected in the AI capabilities that everyone is closely concerned about—

Each GPU core in the M6 directly integrates a Neural Accelerator, paired with a Dual 16-core Neural Engine.

In other words, the GPU, originally designed for graphics and general-purpose parallel computing, has now grown an additional module specifically for AI computation!

Looking at this set of parameters alone, it might not be immediately clear—but when applied to a Mac mini, the data becomes much more tangible—

Compared to the M4 Mac mini, the M6 version offers up to 4x higher AI performance, up to 2x faster GPU, and up to 40% faster CPU.

When running LLMs with LM Studio, prompt processing performance reaches up to 4.8 times that of the M4 and up to 13.5 times faster than the first-generation M1 Mac mini.

Moreover, the unified memory bandwidth has now reached 170 GB/s, approximately 10% higher than the M5.

Although the base model still comes with 16GB, it can be upgraded to a maximum of 32GB, making the throughput clearly better suited for running small- to medium-sized models, persistent agents, and AI workflows locally.

You can also tell that Apple’s perspective on the Mac mini has shifted a bit...

Apple itself directly mentioned agentic AI workflows and always-on agentic computing in its press release.

Johny Srouji specifically mentioned that the Mac mini can serve as a home computer, a professional workstation, or even as an always-on agentic device.

The small box that once sat quietly behind a monitor, costing around $600, has now been pushed by Apple toward the direction of a desktop AI agent server.

The model can run locally, the agent can stay running continuously, and code, files, and tasks can all remain on your local machine.

Small as it is, it’s already doing work that’s increasingly like a server.

Of course, the value of this little box has also undergone a leap...

The 2024 M4 Mac mini was initially priced at just 4,499 RMB in China, but the M6 version now starts at 6,999 RMB.

Over two years, it became 2,500 yuan more expensive, with a price increase of over 55%.

After AI capabilities took off, the Mac mini’s small and beautiful price is starting to seem less small. (doge)

Then comes the real powerhouse—the M5 Ultra.

If the M6 is giving regular Macs a powerful AI performance boost, then with the M5 Ultra, Apple is squarely targeting desktop AI workstations.

Notably, this is the first time Apple has implemented a quad-die architecture in its M-series SoCs.

Let me briefly explain what a four-die architecture is—

From an official technical perspective, it connects two sets of dual-die M5 Max chips via the new-generation UltraFusion technology to form a quad-die system—not simply stacking four chips like layers of a pastry.

Performance is also impressive, featuring up to 36 CPU cores, 80 GPU cores, 32 neural engine cores, 512GB of unified memory, and 1.2TB/s memory bandwidth.

AI peak performance reaches up to 4.3 times that of the M3 Ultra and 9.8 times that of the M1 Ultra; large model prompt processing in LM Studio is up to 4 times faster than on the M3 Ultra.

Another very Apple-like and highly suitable feature for local AI is unified memory.

On traditional PCs, CPU memory and GPU memory operate independently; trying to provide hundreds of gigabytes of GPU memory would be extremely costly and engineeringly challenging.

But! The Mac Studio's 512GB unified memory can be directly shared among the CPU, GPU, and other computing units.

This means that many large models, which previously required servers or multiple high-end GPUs to run, can now run directly on a single desktop machine—as long as the model weights fit.

Even one seems too few—you can keep chaining them together...

During the WWDC demo, the official team even used a massive model as an example, splitting its weights across multiple Mac Studio units to run the model when it couldn't fit into a single machine with 512GB of memory.

Under a four-node setup, distributed AI inference performance can reach up to three times that of a single machine.

So by now, the direction of this generation of Mac Studio is quite clear—

Apple is integrating Apple Silicon, massive unified memory, MLX, and Thunderbolt RDMA into a complete local AI computing stack.

512GB per unit? If it's too small, keep linking units together.

Now this is great—while others are still selling you AI-powered computers, Apple is starting to sell you desktop AI data centers??

Questions also arose.

Such powerful unified memory just happened to collide with one of the most sought-after components in the AI industry over the past two years: memory.

The Wall Street Journal mentioned that previous generations of the Mac mini and Mac Studio have quietly become hits among AI developers.

Especially after demand rose for local large models, Coding Agents, OpenClaw, and similar applications, the advantages of high-memory Macs were suddenly rediscovered—

After purchasing the machine, you can run the model repeatedly without counting tokens for each invocation.

AI companies are competing for memory, data centers are competing for memory, and PC manufacturers are competing for memory.

Now even people running Agents on Macs are coming to compete for memory—supply and demand have directly collided...

The previous generation's highest memory configuration for Apple was once even removed from the website, and although the new Mac Studio has reinstated the 512GB unified memory option, Apple has already stated in advance—

This version will be available at the end of October, while the other Mac Studio models launched on September 22.

Moreover, the prices didn't hold back either: the previous-generation Mac Studio with M4 Max started at CNY 16,499 in China, while the M3 Ultra version started at CNY 32,999.

This generation starts at ¥19,999 for the M5 Max and ¥46,999 for the M5 Ultra.

In other words, the Ultra entry-level model is now 14,000 yuan more expensive.

The price surge has truly gone hand in hand with AI computing power...

Interestingly, the person who played this card is also preparing to step down from the position of Apple’s CEO.

On August 23, Apple had just thrown a farewell party for Cook.

On September 1, John Ternus officially took over.

So, from a timeline perspective, these two chips do carry a bit of the flavor of Cook’s “final dance”—

In the final days of the Cook era, Apple stepped into 2nm technology on one hand, while turning the Mac into a 512GB desktop AI powerhouse.

How could this not be the best retirement gift you could give yourself?

Of course, the foldable iPhone set to be released next month, although unveiled by the new CEO, is theoretically still credited to Cook.

Jobs pioneered the iPhone, and then Cook... bent it... folded it...

Reference link:

[1]https://www.apple.com/newsroom/2026/08/apple-introduces-m6-and-m5-ultra-for-a-big-leap-in-performance-and-ai-compute/

[2]https://www.apple.com/newsroom/2026/08/apple-unveils-a-more-powerful-mac-mini-featuring-the-all-new-m6-and-m5-pro/

This article is from the WeChat public account "Quantum Bit," authored by Meng Yao.