Tuesday, October 31, 2023

Apple M3, M3 Pro, and M3 Max

Apple (Hacker News):

These are the first personal computer chips built using the industry-leading 3-nanometer process technology, allowing more transistors to be packed into a smaller space and improving speed and efficiency.


The M3 family of chips features a next-generation GPU that represents the biggest leap forward in graphics architecture ever for Apple silicon. The GPU is faster and more efficient, and introduces a new technology called Dynamic Caching, while bringing new rendering features like hardware-accelerated ray tracing and mesh shading to Mac for the first time. Rendering speeds are now up to 2.5x faster than on the M1 family of chips. The CPU performance cores and efficiency cores are 30 percent and 50 percent faster than those in M1, respectively, and the Neural Engine is 60 percent faster than the Neural Engine in the M1 family of chips. And, a new media engine now includes support for AV1 decode, providing more efficient and high-quality video experiences from streaming services.


Additionally, support for up to 128GB of memory unlocks workflows previously not possible on a laptop, such as AI developers working with even larger transformer models with billions of parameters.

Tim Hardwick:

However, looking at Apple’s own hardware specifications, the M3 Pro system on a chip (SoC) features 150GB/s memory bandwidth, compared to 200GB/s on the earlier M1 Pro and M2 Pro. As for the M3 Max, Apple says it is capable of “up to 400GB/s.”


Notably, Apple has also changed the core ratios of the higher-tier M3 Pro chip compared to its direct predecessor. The M3 Pro with 12-core CPU has 6 performance cores (versus 8 performance cores on the 12-core M2 Pro) and 6 efficiency cores (versus 4 efficiency cores on the 12-core M2 Pro), while the GPU has 18 cores (versus 19 on the equivalent M2 Pro chip).


According to Apple, the M3 Neural Engine is capable of 18 TOPS, whereas the A17 Pro Neural Engine is capable of 35 TOPS.


Taken together, it’s presently unclear what real-world difference these changes make to M3 performance when pitted against Apple’s equivalent precursor chips in various usage scenarios[…]

Phil Dennis-Jordan:

So the M3 Pro is basically a 50% scaled-up M3: unlike the M2 Pro it doesn’t have double the memory channels, “only” half again.

Jeff C.:

It seems that they’re doing a bit more this generation to differentiate between Pro and Max.

Previously, the Pro and Max had the same number of CPU cores, so I never had any interest in the Max once I realized that my work fit into the Pro’s RAM ceiling. Now, the Pro’s CPU core advantage over the base chip has been cut in half. To get double the cores of the M3 you need the M3 Max.

Om Malik:

The new M3 chips are coming at an opportune time — Apple’s rivals, Qualcomm, Nvidia, AMD, and Intel, have been making noises about catching up with Apple. […] Qualcomm recently announced the Snapdragon X, a PC chip that it says is better than the M2 processor. Nvidia, too, is working on its own chip, as is AMD. All three companies are using Arm’s technology. Intel, on the other hand, is moving forward with its own technologies.


How caching is implemented varies based on the intended use — whether it be for gaming, professional graphics, or data center applications. NVIDIA, for example, employs various forms of cache, including L1/L2 caches and shared memory, which are dynamically managed to optimize performance and efficiency. AMD uses large L3 caches (“Infinity Cache”) to boost bandwidth and reduce latency — an approach beneficial for gaming. Intel’s Xe graphics architecture focuses on smart caching, balancing power efficiency and performance.


Apple has a substantial opportunity to integrate generative AI into its core platform, mainly because of its chip and hardware-level integration.


Update (2024-05-09): See also:

3 Comments RSS · Twitter · Mastodon

Note that the M3 Pro in the $1999 14” MacBook Pro — the standard SKU/price point that’s highlighted in the marketing — only has 5 performance cores. The 6th performance core costs an extra $200 as a BTO option.

Between this, the price hike on the M2 MacBook Air vs the M1, and the M1 Air still not getting any cheaper three years after release, Apple must be struggling with reality of higher silicon costs for these advanced nodes. I guess I was wrong in my assumption early on that switching to ARM would allow Apple to create more affordable entries into the Mac ecosystem.

Given that CPU workloads only seem to be getting a very modest bump from the M3 — and at the cost of higher peak power, no less — Apple's going to face an uphill battle to convince customers to upgrade. It’ll probably only get worse, as these effective price hikes come at a time when memory and storage prices are likely at historical lows for Apple.


The BTO part of it is nothing new, though. And I'm not sure it's so much cost as it is yield, especially this time around, with TSMC struggling with 3nm. (Of course, this approach is win-win for Apple: you can still sell the product, and for those who care about the extra core, you can ask a big markup.)

But, for example, the M1 Air was originally available with either 7 or 8 GPU cores. The M2 Air is currently available with 8 or 10. The 14-inch M1 Pro started at 6/2 CPU cores; 8/2 was an upgrade. The M2 Pro changed that to 6/4 or 8/4.

What's new here is that the M3 Pro's layout seems to be different from both the M3 (that's not new) but also from the M3 Max (whereas the M1 Pro and M2 Pro were chopped versions of the M1 Max and M2 Max), and as a result, the core counts, memory bandwidths, etc. are quite different.

How much of this is Apple deliberately doing segmentation (I sure hope not; $3,200 is too high a starting point for the Max) vs. yield issues over at TSMC remains to be seen with the M4 series, I guess.

Jeff C is right. People wanting more CPU and RAM are being not-so-subtly strong-armed into the Max, with a clearer and wider differentiation from the Pro. Shame; I don't need insane GPU. Guess I'll have to scale back my aspirations.

Leave a Comment