Apple’s M3 processor family is here. And it promises. Those from Cupertino have presented three chips with different features and a common microarchitecture: the M3, M3 Pro and M3 Max CPUs. At the moment there is no sign of a very likely M3 Ultra processor, although we can be reasonably sure that it will arrive in the future, possibly when Apple decides to renew its Mac Pro and Mac Studio desktop computer families.
These are the first microprocessors for laptop and desktop computers manufactured using TSMC’s 3 nm photolithography, which is currently the most advanced in the large-scale production phase of this Taiwanese company. However, the microarchitecture of these chips also introduces several very important improvements that, on paper, should allow them to clearly outperform the equivalent M2 family CPUs.
The subsystem that incorporates the most relevant improvements is the graph. For the first time a Mac processor implements hardware acceleration for ray tracing, and it’s also the first time the integrated GPU relies on a dynamic cache. In addition, the performance of the CPU cores and the Neural Engine for artificial intelligence is, according to Apple, much higher than in the M2 processors, and the M3 chips are designed to coexist with a larger unified memory map (up to 128 GB if we stick to the M3 Max CPU).
The Apple M3, M3 Pro and M3 Max microprocessors, in figures
|
m3 max
|
m3 pro
|
m3
|
M2
|
M1
|
|
|---|---|---|---|---|---|
|
photolithography
|
3nm
|
3nm
|
3nm
|
5nm (2nd generation)
|
5nm
|
|
number of transistors
|
92 billion
|
37 billion
|
25 billion
|
20 billion
|
16 billion
|
|
maker
|
TSMC
|
TSMC
|
TSMC
|
TSMC
|
TSMC
|
|
number of cpu cores |
Up to 16 |
Up to 12 |
8 |
8 |
8 |
|
high performance cores (ar) |
12 |
6 |
4 |
4 |
4 |
|
high efficiency (ae) cores |
4 |
6 |
4 |
4 |
4 |
|
number of graphics cores |
Up to 40 |
Up to 18 |
10 |
10 |
8 |
|
NEURAL ENGINE CORE (NE) |
16 |
16 |
16 |
16 |
16 |
|
maximum unified memory map |
128GB |
36GB |
24GB |
24GB |
16 GB |
|
main memory technology |
LPDDR5-6400 |
LPDDR5-6400 |
LPDDR5-6400 |
LPDDR5-6400 |
LPDDR4X-4266 |
|
unified memory bandwidth |
Up to 400 GB/s |
150GB/s |
100GB/s |
100GB/s |
68GB/s |
|
video encoding and decoding |
Hardware accelerated H.264, HEVC, H.265, ProRes, ProRes RAW and AV1 Hardware accelerated H.264, HEVC, H.265, ProRes, ProRes RAW and AV1 |
Hardware accelerated H.264, HEVC, H.265, ProRes, ProRes RAW and AV1v |
Hardware accelerated H.264, HEVC, H.265, ProRes, ProRes RAW and AV1 |
8K H.264, H.265, ProRes and ProRes RAW |
4K H.264 and H.265 |
|
connectivity |
3 x Thunderbolt 4/USB 4 |
3 x Thunderbolt 4/USB 4 |
2 x Thunderbolt 3/USB 4 |
2 x Thunderbolt 3/USB 4 |
2 x Thunderbolt 3/USB 4 |
This is the performance of the M3 processors that Apple promises us
The configuration of the cores of the M3 family chips is consistent with what the M1 and M2 series have proposed to us. The entry processor to the family, the M3 chip, incorporates 8 CPU cores (4 high performance and another 4 high efficiency), and 10 GPU cores. The M3 Pro processor is available in two different versions– with 11 CPU cores and 14 GPU cores, or with 12 CPU cores (6 high-performance cores and 6 high-efficiency cores) and 18 GPU cores. Finally, the M3 Max chip brings together up to 16 CPU cores (12 high-performance cores and 4 high-efficiency cores) and up to 40 graphics cores.
Apple has told us what we can expect about the performance of its new processors
As on other occasions, Apple has only revealed some details about the architecture of the M3 processors. As I mentioned a few lines above, the most relevant improvements come from the graphic logic (we will investigate it in the next section of this article), although yes, those from Cupertino have told us what we can expect about the performance of your new processors by putting them to the test with video editing, image processing, code compilation or productivity applications.
The most prudent thing is that we take the performance indices that the brands give us with some reluctance because it is evident that they are an interested party, but they can help us to get a rough idea about the productivity of the new chips. Of course, as soon as the new MacBook Pro or iMac equipped with an M3 chip fall into our hands we will analyze their performance in depth. In the graph below we can see that, according to Apple, the M3 processor is noticeably faster than the M2 and M1 chips when scaling images with Photomator.
When rendering images using Redshift, the M3 Pro chip far outperforms the M2 Pro and M1 Pro processors than the M3 did over its predecessors in the previous test. The higher clock frequency at which the CPU and GPU cores of the M3 Pro processor presumably work works in its favor, although it is very likely that Apple engineers have also introduced important optimizations in the microarchitecture of the M3 chips compared to the implementation of M2 processors.
On its website Apple has published many more performance tests, but the three we have selected illustrate quite well what productivity the M3 chips promise us. In the following graph we can see that the M3 Max, the most capable until the M3 Ultra processor arrives in the future, is much faster when rendering images in Redshift than the M2 Max and M1 Max chips.
This improvement is a priori the result of the increase in the clock frequency and the optimizations introduced by Apple in the microarchitecture of the M3 processors. Of course, TSMC’s 3nm lithography in theory works in its favor not only if we stick to its performance per watt; It also determines the maximum clock frequencies at which the CPU cores are capable of working.
According to Apple, the high-performance cores of the M3 processors are 30% faster than the comparable cores of the M1 chips. And the high-efficiency cores are, again according to Cupertino, 50% faster than the comparable cores of the M1 chips.
Unfortunately, Apple has barely given us a couple of hints about the efficiency of the M3 microprocessors, although it has promised us something revealing: its performance in multithreaded usage scenarios is the same as that offered by comparable M1 chips, but your energy consumption is reduced by half. In this area, 3 nm lithography makes the difference.
Hardware ray tracing and dynamic cache come to the Mac GPU
Apple assures that the graphics of the M3 processors play in another league. Could be. We will know for sure when we have the opportunity to thoroughly analyze the first Macs equipped with these chips, but the truth is that, on paper, the improvements implemented in its graphic logic look very good. One of the most important is dynamic cache, a work strategy that allows the GPU to decide in real time how much local memory it needs to reserve for each task. This technique in theory allows you to maximize the use of graphical logic and its performance. At the same time it optimizes the use of local memory.
The integrated GPU in the M3 processors implements hardware specifically dedicated to ray tracing acceleration
Additionally, the integrated GPU in the M3 processors implements hardware specifically dedicated to ray tracing acceleration. PC gamers know well that this rendering strategy is much more demanding on graphics hardware than traditional rasterization, so the presence of dedicated hardware is welcome. According to Apple, the performance of M3 chips is up to 2.5 times greater than that of M1 processors when the . Sounds good.
However, the graphics logic of the M3 processors implements one more improvement that is worth not overlooking: hardware acceleration of the . Broadly speaking, this technique acts on the geometry of the scene with the purpose of transforming complex geometry into a package of simpler meshes that can be rendered with much less effort. According to Apple, the GPU of the M3 processors is capable of delivering the same performance as the graphics logic of the M1 chips, but consuming half the energy.
Up to 128 GB of unified memory
The implementation of a unified memory map that is accessed by both the CPU cores and the GPU responds to the search for a strategy that allows reducing latency, increasing energy efficiency, and at the same time, improving transfer speed. This technique is still present in the M3 processors, but these chips can work in tandem with a larger unified memory map. The M3 processor maintains the maximum 24 GB of the M2 and M1 chips, but the M3 Pro increases this figure to 36 GB, and the M3 Max to a quite impressive 128 GB.
