Apple introduces new M3 chip lineup, starting with the M3, M3 Pro, and M3 Max [Updated]

NEW YORK—None of the new Macs that Apple is announcing at its "Scary Fast" product event today look very different from the ones they're replacing on the outside, but the inside is another story. This is the first batch of Macs to include Apple's next-generation M3-series chips, and unlike past years, Apple is introducing multiple M3 performance tiers all at the same time.

The M3, M3 Pro, and M3 Max all share the same underlying CPU and GPU architectures, the same ones used in the iPhone 15 Pro's A17 Pro chip. Also like the A17 Pro, all M3 chips are manufactured using a new 3 nm process from Taiwan Semiconductor (TSMC). Let's dive into everything we know about the M3 family's capabilities, plus the differences between each performance tier.

Meet the Apple M3 family

Apple says that the performance cores in any given M3 processor can run up to 30 percent faster than the M1's performance cores, and that the efficiency cores are up to 50 percent faster. Most of Apple's direct performance comparisons were to the M1 generation, which is useful insofar as M2 Mac owners aren't likely to want to spring for M3, but it has the added marketing benefit of making the performance increases sound larger than they are.

To put those numbers in context, the M2’s P-cores were already between 10 and 20 percent faster than the M1’s, depending on the chip. Isolating the E-cores in benchmarks is a bit more difficult, but Apple says that the M3 is up to 35 percent faster than the M1 at peak power. Apple said the M2 was about 18 percent faster than M1.

The M3’s GPU is also based on the same architecture as the A17 Pro, which means that all M3 Macs are picking up hardware-accelerated ray-tracing support like what you find in more recent Nvidia GeForce and AMD Radeon GPUs. As in desktop PCs, enabling ray tracing in most games and apps is going to come with a performance penalty, which may make it more usable on the M3 Pro and Max than it is in the basic M3, but every GPU in the family at least supports the capability.

Other hardware features added to the M3 GPU include "Dynamic Caching," which claims to allocate "only the exact amount of memory needed... for a given task," allowing more tasks to utilize the GPU at a given time. Like DirectX 12 Ultimate GPUs, the M3 GPU also supports hardware-accelerated mesh shading, useful for rendering scenes that use a whole lot of small objects.

Aside from new hardware features, Apple says that peak GPU performance for the M3 series is up to 65 percent faster than M1's, though as with the CPU performance this will likely vary based on which chips you're comparing. In our tests, the various M2-series GPUs are generally between 25 and 30 percent faster than their M1 equivalents, so the year-over-year performance increases from M1 to M2 to M3 are fairly even. Apple says that the M3's rendering performance is up to 2.5 times faster than the M1 for workloads that take advantage of the new hardware acceleration features.

Finally, the media engine in the M3 series supports hardware-accelerated decoding for AV1 video streams. Newer PC GPUs also support hardware-accelerated encoding, but it’s still a fairly niche feature with relatively limited usefulness for home users beyond high-resolution video broadcasting. Still, it's something that most modern integrated PC GPUs support (or will support very soon), so it's too bad not to see it here.

M3, M3 Pro, M3 Max

Compared to the M2, the plain-old M3 doesn’t get any additional cores, so it will rely solely on architectural improvements and clock speed bumps to increase performance—and it has 25 billion transistors, 5 billion more than M2, so there's still quite a bit of new hardware here. It’s still an 8-core CPU, split evenly between performance and efficiency cores, and a 10-core GPU (with a partially disabled 8-core GPU in some entry-level models). Integrated on-package system memory starts at 8GB and maxes out at 24GB, same as the M2.

The M3 Pro is a 37-billion transistor chip, which is 3 billion fewer than the M2 Pro. That makes sense once you look at the core counts; M2 Pro had eight performance cores and four efficiency cores, plus as many as 19 GPU cores. The M3 Pro still has 12 cores, but it's split evenly between six performance cores and six efficiency cores, and the GPU tops out at 18 GPU cores. Memory bandwidth is also somewhat lower than the M2 Pro, at 150GB/s instead of 200GB/s. Maximum memory capacity does increase slightly, from 32GB to 36GB, while the minimum memory capacity goes up from 16GB to 18GB.

I still expect The M3 Pro to be an upgrade over the M2 Pro because of the updated architectures involved, but it looks like less of an upgrade than the M3 (which keeps core counts the same) or the M3 Max (which increases them). The M2 Pro and M2 Max used the same CPU core configuration, and my best guess is that the company wanted to create more of an incentive to jump from Pro to Max for people who don't care about GPU performance.

As for the M3 Max, the 92 billion transistor count makes a huge jump from the M2 Max's 67 billion. A lot of that is accounted for by the CPU M3 Max, which includes 12 performance cores and four efficiency cores, four more performance cores than the M2 Max. The GPU also gets a little bigger, jumping from a max of 38 cores to a max of 40. Maximum memory capacity also goes up from 96GB to 128GB, and minimum memory capacity goes from 32GB to 36GB.

All three of the M3 processors have two different sub-variants, one "binned" version with some CPU and/or GPU cores disabled and a fully enabled version. The M3 Max supports different amounts of memory based on which version you buy; the entry-level version can come with 36 or 96GB of RAM, while the fully-enabled version starts at 48GB and can be configured with 64GB or 128GB. The entry-level version also has a slightly lower memory bandwidth of 300GB/s, compared to the fully-enabled 400GB/s. Previous Max chips always got the maximum 400GB/s no matter how they were configured, so the performance of the lower-end Max could be a little lower than its core counts suggest, particularly for graphics rendering.

Aside from core counts, display support is still a noteworthy differentiator. The low-end M3 still supports a total of two screens, counting the internal display, so M3 systems with built-in screens like the cheapest 14-inch MacBook Pro, the iMac, and any forthcoming M3 MacBook Airs will still only support a single external display. The M3 Pro supports a total of three displays, and the M3 Max can drive a total of five.

The M3 lineup also features the same number of ProRes video encoding and decoding engines, for those who use them—one in the M3 and M3 Pro, and two in the M3 Max.

The only member of the M3 family that Apple didn't talk about today, assuming there are no new family members coming, was the M3 Ultra. This top-end M3 chip will presumably just be two M3 Max chips strapped together, like the M1 Ultra and M2 Ultra before it. But Apple just refreshed the Mac Studio and Mac Pro with the M2 Ultra in June, so we wouldn't expect to see replacements before summer 2024 at the earliest.

Update, 10/31/2023: We've added additional information about each M3 chip's memory bandwidth, all possible memory configurations, and different CPU and GPU core counts for each version of each processor.