Take a fresh look at your lifestyle.

The Microarchitecture of Apple’s M1 Pro and M1 Max Chips Explained: Their Secret Weapon Is Scalability That Hasn’t Said Its Last Word (Not By Far)

Apple has lived up to expectations. We knew that your first MacBook Pros equipped with an M1 microprocessor They were close, and yes, they are about to come out of the oven. But they will not arrive with the same M1 chip that we can find since the end of last year in the MacBook Air, and what a good taste in our mouth it left us after thoroughly testing it to find out if its performance and consumption were really so competitive as their creators claimed.

The new MacBook Pros will be equipped with the M1 Pro and M1 Max microprocessors, two versions of the M1 chip that maintain the same microarchitecture of the original model, but which, at the same time, are significantly more ambitious This is due to the way in which they implement the three fundamental subsystems of these SoCs: the CPU, the GPU and the unified memory that acts as a meeting point for the main functional elements of these chips.

The M1 Pro and M1 Max processors have been sized to fit the usage scenarios for which MacBook Pros are designed.

Not surprisingly, Apple has beefed up its M1 chip to make it fit well in the usage scenarios for which MacBook Pros are indicated, which are primarily designed to measure up to content creation applications that can push both the CPU and GPU to the limit. But, despite the fact that we expected one or more more ambitious variants of the original M1, the capabilities that the M1 Pro and M1 Max processors have on paper are impressive.

We won’t know for sure if your overall productivity and performance / watt ratio are as good as Apple tells us until we get them down to business and in harsh conditions, something we hope to be able to do as soon as possible. However, in the meantime we can inquire into its architecture, because, as we are about to verify, the engineers who have designed these chips have worked hard to size them even one step beyond what we could foresee if we took as a starting point the original M1 processor.

Apple M1, M1 Pro and M1 Max: technical specifications

M1

M1 Pro

M1 max

microarquitectura

M1

M1

M1

photolithography

5 nm

5 nm

5 nm

transistors

16 billion

33.7 billion

57 billion

cores (cpu)

8 cores

Up to 10 cores

10 cores

high performance cpu cores

4 cores

Up to 8 cores

8 cores

Instruction cache (L1)

192 KB

192 KB

192 KB

data cache (L1)

128 KB

128 KB

128 KB

level 2 cache (l2)

12 MB

24 MB

24 MB

high efficiency cpu cores

4 cores

Up to 2 cores

2 cores

instruction cache (l1)

128 KB

128 KB

128 KB

data cache (l1)

64 KB

64 KB

64 KB

level 2 cache (L2)

4 MB

4 MB

4 MB

unified memory

Up to 16GB

Up to 32GB

Up to 64GB

MEMORY TECHNOLOGY

LPDDR4X

LPDDR5

LPDDR5

Interface

128 bits

256 bits

512 bits

MEMORY BANDWIDTH

68,25 GB/s

Up to 200GB / s

Up to 400GB / s

neural engine (NE)

16 cores

16 cores

16 cores

yield (NE)

11 trillion operations per second

11 trillion operations per second

11 trillion operations per second

NUCLEOS (GPU)

Up to 8 cores

Up to 16 cores

Up to 32 cores

execution units (GPU)

128

2048

4096

SHADER TFLOPS (FP32)

2,6

5,2

10,4

texture rate

82 Gteels / s

164 Gteels / s

327 Gteels / s

pixel rate

41 Gpíxeles / s

82 Gpíxeles / s

164 Gpíxeles / s

codecs

H.264, HEVC y ProRes

H.264, HEVC y ProRes

H.264, HEVC y ProRes

external monitors

1

2

4

Scalability is the true backbone of the M1 microarchitecture

The integration technology used by TSMC to produce the M1 Pro and M1 Max processors is the same as the one used in the manufacture of the original M1 chips: FinFET photolithography (Fin Field-Effect Transistor) de 5 nm. Of course, both the Pro review and the more ambitious Max incorporate many more transistors than the first M1 chip: 33.7 billion for the M1 Pro and no less than 57 billion for the M1 Max.

The Pro version of the SoC M1 can incorporate a maximum of 10 cores for the CPU, so that up to 8 of them are high-performance and the remaining 2 high-efficiency. The purpose of this hybrid concept inspired by ARM’s big.LITTLE architecture processors is balance performance and consumption to increase the performance / watt ratio of these chips as much as possible.

The hybrid architecture of the M1 processors seeks to balance the performance and consumption of these chips to the maximum

The M1 Max processor, however, is only available with 10 CPU cores divided into 8 high-performance and 2 high-efficiency. Of course, the cache memory subsystem of the M1, M1 Pro and M1 Max processors is identical. Each of the high-performance cores incorporates a 192 KB instruction cache and a 128 KB data cache. In the M1 chips, the shared level 2 cache has a capacity of 12 MB, while in the M1 Pro and M1 Max it goes up to 24 MB.

The strategy Apple has followed with high-efficiency cores is essentially the same. Each of them incorporates a 128 KB instruction cache and a 64 KB data cache. Yes, this time shared level 2 cache is identical on the M1, M1 Pro and M1 Max chips because in all three it has a capacity of 4 MB.

M1pro
M1pro

One of the most important decisions made by the engineers who have designed the M1 chips is to choose a unified and shared memory map accessible by both the CPU and GPU and the Neural Engine machine learning engine. The original M1 processor incorporates a maximum of 16 GB LPDDR4X, while the M1 Pro and M1 Max revisions extend the unified memory up to 32 and 64 GB respectively. One more note: the memory of these last two chips is of the LPDDR5 type.

The unified memory interfaces of the M1 Pro and M1 Max chips have a ‘width’ of 256 and 512 bits respectively

However, there is one more fact that we cannot ignore. The memory interface of the M1 processor is 128-bit ‘wide’, while the interfaces of the M1 Pro and M1 Max chips extend to 256 and 512 bits respectively. Not surprisingly, memory interface characteristics have a direct impact on bandwidth, slightly exceeding 200GB / s on the M1 Pro processor and flirting with 400 GB / s on the M1 Max.

On the other hand, the Neural Engine logic, which is the name that Apple has given to the machine learning engine that it has implemented in these microprocessors, is identical in all three chips, so in this field the SoC M1 Pro and M1 Max do not They outperform the original M1. All three have 16 cores dedicated to running artificial intelligence algorithms and are capable of running up to 11 trillion operations per second. And these billions are ours, not Anglo-Saxons.

M1max
M1max

We still have to investigate another of the essential subsystems of these chips: their graphical logic. The 8 cores that the original M1 processor GPU incorporates at most pale in comparison to the 16 cores that the M1 Pro chip can integrate at most. And, above all, before the maximum 32 cores on which the SoC M1 Max can lean. In fact, if we stick to its brute force on paper, the graphical logic of these last two processors rivals that of the previous generation of dedicated mid / high-end graphics cards. And it is not but bad for integrated graphics.

The performance figures that Apple promises us when it describes the graphics capacity of its M1 chips are very interesting. The M1 Pro processor, always according to its creators, gives us 5.2 TFLOPS (FP32), and the more ambitious M1 Max reaches 10.4 TFLOPS, a figure slightly higher than the 10.28 TFLOPS that a PlayStation 5 console manages to deliver when fully used. This is how the little beast that is the M1 Max chip spends them in theory. If you are curious in the specifications table that we publish above, you can check the texture and pixel rates of the graphical logic of these SoCs.

The energy efficiency of the new M1 Pro and M1 Max, according to Apple

Everything that we are going to see in this section of the article we must take with a lot of reservations because it is the information that Apple gives us to describe the performance of its new processors when measured with PC chips. And obviously, Apple is a stakeholder. When we thoroughly analyze the new M1 Pro and M1 Max we will draw our own conclusions, but it is not bad to open our mouths by looking at what the Cupertino people promise us.

According to Apple, its most powerful M1 processors match the performance of a next-generation x86-64 chip for laptops, consuming 70% less

According to them, as we can see, their more powerful M1 processors match the performance of a next generation x86-64 chip for laptops, and they do it. consuming 70% less. Apple also claims that when fully used, its SoCs are 1.7 times faster. It is a pity that you do not state in full transparency what the complete configuration of the PC you have used in your tests was because exposing the data in this way raises reasonable suspicions. Of course, at least he confesses to having compared his MacBook Pros with M1 Pro and M1 Max chip with an MSI GP66 Leopard laptop, which presumably incorporates an Intel Core i7-11800H CPU.

M1 performance
M1 performance

On the next slide, Apple tells us that its 16-inch MacBook Pro with M1 Pro processor is at full power. consume 70% less than a laptop with a dedicated GPU. In this case, it seems, they have compared it to a Legion 5 from Lenovo, which presumably had to incorporate a Ryzen 7 5800H processor from AMD and a GeForce RTX 3060 GPU from NVIDIA.

M1consumo
M1consumo

On the last slide Apple claims that its MacBook Pro with an M1 Max processor performs better and consumes 40% less than a compact professional notebook PC with a dedicated GPU. And also that it performs essentially the same as a high-end notebook PC, but consuming 100 watts less. Apparently in these tests Apple has used an MSI GE76 Raider laptop and a Razer Blade 15 Advanced.

Performance m1max
Performance m1max

The clues we have draw the future of M1 chips, and it is a promising future

There is no better way to face the final stretch of this article than trying to intuit what the next microprocessors of the M1 family will offer us. What we have seen so far clearly reflects that for Apple it is relatively simple scale this architecture in order to drastically increase the performance of its microprocessors. This strategy is very similar to the one Intel is going to implement with its Alder Lake hybrid architecture microprocessors.

The M1 chips for the more ambitious iMac and Mac Pro will have even more CPU and GPU cores. It’s not official, but we can trust it

We can be sure of one thing: when Apple introduces its M1 chip into its more advanced desktop computers (24-inch iMac already incorporate it) we will see that there will be versions that will have more cores of all kinds than even the powerful M1 Max. More CPU cores. More GPU cores. And maybe even a larger unified memory.

To make this possible, Apple will not need TSMC to use more advanced integration technology, but when it is available, and this moment will not be long in coming, the M1 chips will be able to deliver. another leap forward. There is no doubt that a very interesting time is coming for microprocessor enthusiasts during which the competition between Apple, Intel and AMD will be more aggressive than ever. So be it because, once again, we, the users, will benefit the most.