M4 Deep Dive - Apple's First Armv9 Processor

Summary

Apple's new M4 processor, introduced in the iPad Pro, boasts up to 10 cores with enhanced performance and efficiency cores, featuring improved branch prediction and deeper execution engines. The M4 integrates Next-Generation ML accelerators within its CPU, supporting ARMv9.4 architecture, marking Apple's first venture into this territory. Geekbench scores indicate a significant leap in performance for the M4, especially when compared to its predecessors. The device also includes a 10-core GPU with dynamic caching and hardware-accelerated ray tracing. Speculations on future M4 variants and its potential integration into iPhones are also discussed.

Takeaways

• 🍏 Apple has introduced the new M4 processor, which is initially being used in an iPad Pro rather than a MacBook or MacBook Air.
• 🔍 The M4 processor features up to 10 cores, with configurations of up to four performance cores and six efficiency cores, an upgrade from the M3's 4+4 setup.
• 🚀 Enhanced features of the M4 include improved branch prediction, wider decode and execution for performance cores, and a deeper execution engine for efficiency cores.
• 🌟 A significant highlight is the inclusion of Next Generation ML accelerators within the CPU, which is crucial for tasks involving machine learning, computer vision, and scientific simulations.
• ⚠️ Consumers should be aware that not all M4 processors come with 10 cores; some may have nine, and this isn't always clear in Apple's marketing materials.
• 💾 The 10-core version of the M4 is only available with 1 terabyte of storage and 16 gigabytes of RAM, which may be a limitation for some users.
• 🛠️ The M4's microarchitecture has been improved with features like scalable Matrix extensions, which are part of the ARM v9.4 architecture and not available in ARM v8.
• 📊 Geekbench scores indicate that the M4, especially the 10-core version, shows a significant leap in performance over previous generations like the M1, M2, and M3.
• ⏱️ The M4 operates at a boosted clock speed of 4.4 GHz, which contributes to its improved performance compared to its predecessors.
• 📈 When comparing the M4 to competitors like the Snapdragon X+ and X Elite, the M4 shows promising performance, but actual power efficiency and real-world testing will provide a clearer picture.
• 🎮 The M4 also includes a new 10-core GPU, which builds on the M3's graphics architecture and includes features like dynamic caching and hardware-accelerated ray tracing.

Q & A

What is the new M4 processor from Apple used in?

The new M4 processor from Apple is used in an iPad Pro, rather than a MacBook, MacBook Air, or similar devices.

What is the core configuration of the M4 processor?

The M4 processor has up to 10 cores, consisting of up to four performance cores and six efficiency cores, which is an upgrade from the M3's 4+4 core configuration.

What is the significance of the enhanced Next Generation ML accelerators in the M4 processor?

The enhanced Next Generation ML accelerators in the M4 processor are crucial as they facilitate hardware-accelerated matrix operations, beneficial for scientific simulations, computer vision, and machine learning.

How does the M4 processor's core configuration affect its availability in different iPad Pro models?

The M4 processor with a 10-core configuration is only available in the iPad Pro model with one terabyte of storage and 16 gigabytes of RAM. The nine-core version is available in models with less storage and RAM.

What is the clock speed of both the nine-core and ten-core versions of the M4 processor?

Both the nine-core and ten-core versions of the M4 processor have a boosted clock speed of 4.4 GHz.

What is the difference in microarchitecture between the M1, M2, M3, and M4 processors?

Each processor (M1, M2, M3, M4) has a different internal microarchitecture. The M4 notably includes improved branch prediction, wider decode and execution for performance cores, and a deeper execution engine for efficiency cores, as well as NextGen ML accelerators.

What does the M4 processor's Geekbench scores indicate about its performance compared to its predecessors?

The M4 processor's Geekbench scores show a significant leap in performance, especially when comparing single-threaded scores normalized by clock speed, indicating improved microarchitecture beyond just increased clock speeds.

How does the M4 processor compare to the Snapdragon X+ and Snapdragon X Elite in terms of performance?

While the Snapdragon X+ and X Elite have not yet been tested in laptops, their scores suggest they may outperform the M4 in raw multi-threaded performance, but actual power efficiency and thermal performance will be crucial for real-world comparison.

What new features does the M4 processor's GPU have?

The M4 processor's GPU features a new 10-core design that builds on the M3's architecture, including dynamic caching, hardware-accelerated ray tracing, and mesh shading.

What are some possible future variations of the M4 processor that Apple might introduce?

Possible future variations of the M4 processor could include an M4 Pro with more cores, and an M4 Max with even higher core counts. There is also speculation about the A18 chip for iPhones potentially incorporating Arm v9 architecture.

Outlines

00:00

💻 Introduction to Apple's M4 Processor in iPad Pro

Apple's M4 processor, introduced in the iPad Pro, boasts up to 10 cores, a significant upgrade from the M3's 4+4 core configuration. The new processor includes four performance cores and six efficiency cores, featuring improved branch prediction and a deeper execution engine for enhanced performance. The M4 also integrates next-generation machine learning (ML) accelerators, which are crucial for scientific simulations, computer vision, and machine learning tasks. Apple's presentation highlighted the performance improvements but did not make it clear that the full 10-core configuration is not guaranteed, leading to some customer disappointment. The processor operates at a boosted clock speed of 4.4 GHz and is built using TSMC's second-generation 3-nanometer process node. It's important for potential buyers to understand the specifications and storage options to ensure they receive the desired CPU core count and RAM configuration.

05:01

🔍 Deep Dive into M4's CPU Microarchitecture and Geekbench Scores

The M4's CPU microarchitecture has been redesigned with improved branch prediction, wider decode and execution for performance cores, and a deeper execution engine for efficiency cores. These changes enhance instruction-level parallelism and performance. Apple's first ARM v9 CPU, the M4 includes the scalable Matrix extension, which facilitates hardware-based matrix multiplication, beneficial for ML tasks. Geekbench scores reveal that both the 9-core and 10-core versions of the M4 outperform previous Apple chips like the M3, M2, and M1. However, when normalizing for clock speed, the M4 shows a significant leap in performance over its predecessors, indicating substantial microarchitectural improvements alongside the higher clock speed and new process node technology.

10:01

📊 Comparative Analysis with Snapdragon Processors and M4's GPU

A comparison with Snapdragon X+ and X Elite processors shows that while the M4's 9-core version is slightly behind the Snapdragon X+ in multi-threaded performance, the 10-core version of the M4 is competitive. It's important to note the core configurations differ, with the M4 having a mix of performance and efficiency cores, while Snapdragon processors have all performance cores. The power efficiency of these Snapdragon processors is yet to be determined. Additionally, the M4 features a new 10-core GPU, an upgrade from the M3, with dynamic caching, hardware-accelerated ray tracing, and mesh shading. Predictions for future Apple devices include potential M4 Pro and M4 Max variants with more cores and the possibility of the A18 chip in iPhones incorporating ARM v9 architecture.

Keywords

💡M4 Processor

The M4 Processor is Apple's latest chip, announced for use in the iPad Pro. It signifies a generational upgrade from the M3, featuring enhancements in CPU and GPU architecture. In the script, the M4 is highlighted for its potential impact on device performance, with a focus on its core configuration and clock speed.

💡Core Configuration

Core Configuration refers to the arrangement of processing units within a CPU. The M4 boasts up to 10 cores, a mix of performance and efficiency cores, which is an upgrade from the 4+4 configuration of the M3. The script discusses the significance of this configuration in delivering computational power and efficiency.

💡Branch Prediction

Branch Prediction is a technique used in CPUs to predict the path a program will take, reducing the time wasted on incorrect paths. The script mentions that the M4 has improved branch prediction, which is crucial for enhancing performance by minimizing mispredictions and pipeline stalls.

💡Decode and Execution

Decode and Execution are stages in the CPU pipeline where instructions are translated and then carried out. The script explains that the M4 has wider decode and execution engines for its performance cores, allowing more instructions to be processed simultaneously, which is a key aspect of the chip's improved performance.

💡Efficiency Cores

Efficiency Cores are designed to perform tasks with lower power consumption compared to performance cores. The M4 features six efficiency cores with a deeper execution engine, as mentioned in the script, which suggests that these cores can handle more complex tasks at higher clock speeds while maintaining energy efficiency.

💡Next Generation ML Accelerator

The Next Generation ML Accelerator is a feature within the M4's CPU that enhances machine learning tasks. The script emphasizes the importance of this accelerator, indicating that it allows for hardware-accelerated matrix operations, beneficial for scientific simulations, computer vision, and machine learning applications.

💡Geekbench Scores

Geekbench Scores are benchmark results that measure a processor's performance. The script uses Geekbench scores to compare the M4's single-threaded and multi-threaded performance against its predecessors, the M1, M2, and M3, illustrating the generational improvements in processing power.

💡Clock Speed

Clock Speed refers to the rate at which a CPU can perform instructions and is measured in GHz. The script notes that both the 9-core and 10-core versions of the M4 have a boosted clock speed of 4.4 GHz, which contributes to their performance capabilities.

💡3-nanometer Process Node

A 3-nanometer Process Node is a manufacturing technology used to create smaller, more efficient transistors on a chip. The script mentions that the M4 is built using TSMC's second-generation 3-nanometer process, which allows for more transistors and improved performance and efficiency.

💡Scalable Matrix Extension

Scalable Matrix Extension, specifically version 2, is a feature of ARM v9.4 architecture that enables hardware acceleration for matrix operations. The script explains that this feature is included in the M4's CPU, allowing for more efficient execution of machine learning algorithms.

💡GPU

GPU stands for Graphics Processing Unit, responsible for rendering images, video, and animations. The script describes the M4's new 10-core GPU as an upgrade from the M3, with features like dynamic caching and hardware-accelerated ray tracing, enhancing graphical performance for the iPad.

Highlights

Apple announced the M4 processor, which is notably used in the iPad Pro instead of a MacBook or MacBook Air.

The M4 processor features up to 10 cores, an upgrade from the M3's 4+4 core configuration.

The M4 includes next-generation core features with improved branch prediction and wider decode and execution for performance cores.

Efficiency cores in the M4 have a deeper execution engine, enhancing performance.

Both core types in the M4 feature an enhanced Next Generation ML accelerator, which is crucial for machine learning tasks.

Apple's performance claims are based on Geekbench scores, which show improvements over previous generations.

The M4's core count may vary, with some models having 9 cores instead of 10, which could disappoint some users.

The M4 operates at a boosted clock speed of 4.4 GHz, regardless of having 9 or 10 cores.

The M4 is built using TSMC's second-generation 3-nanometer process node, improving efficiency and performance.

The M4's microarchitecture has been improved for better branch prediction and instruction handling.

The M4's efficiency cores may have higher clock speeds due to a deeper execution engine.

The M4's ML accelerators are part of the CPU and are based on ARM's v9.4 architecture, enabling hardware matrix operations.

Geekbench scores show a significant leap in performance for the M4 compared to its predecessors, even when accounting for clock speed.

The M4's GPU is an upgraded 10-core version from the M3, featuring dynamic caching and hardware-accelerated ray tracing.

The M4's performance puts it in competition with the Snapdragon X+ and X Elite, though power efficiency remains to be seen.

Future M4 variants, such as the M4 Pro or Max, could offer even more cores and enhanced performance.

The A18 in future iPhones might also adopt the M4's ARM v9 architecture, potentially offering an octacore configuration.

Transcripts

so Apple announced its new M4 processor

a couple of weeks ago interestingly

being used in an iPad Pro rather than

appearing in a MacBook MacBook Air or

something like that now I didn't do a

video at the time of the launch because

I didn't want to just repeat the

information that uh Apple were giving in

the presentation you know repeat the

press release I wanted to find out a few

things about this processor I found them

out and today I want to have a look at

them so if you want to find out more

please let me explain

okay so let's go through some of the

highlights of the CPU part of the new M4

processor the M4 has up to 10 core we're

going to talk about that up to 10 core

uh consisting of up to four performance

cores and six efficiency cores so that's

an upgrade from the M3 which was 4 + 4

it's got the next generation core

features including improved Branch

prediction with wider decode and

execution for the performance cores and

a deeper execution engine for the

efficiency cores we'll dive into that

what does all that mean and both types

of cores also feature inhanced Next

Generation ml accelerator now this is a

really really important point and we're

going to dive into that that's why I put

it here in red because we have to

understand that and then Apple are

quoting performance differences and

we'll look at the performance according

to geekbench okay so this is how Apple

presented the uh the new chip so four

performance cores with approved BR

prediction and so on six efficiency CES

giving 10 in total however we have to

notice it's up to 10 core you aren't

guaranteed to get a 10 core CPU and

apple doesn't make it obvious in fact I

know somebody who actually went online

bought the new iPad and then when they

got it with a nine core one they were

pretty disappointed they're actually

going to return it and order the next

one up now the good thing is that both

variants have a boosted clock speed of

4.4 GHz that's the same if you get the

nine core or the 10 core and both built

using tsmc second generation 3 nanometer

process node so what's actually

happening here when you do dig down into

the specifications you don't get shown

that on the page where you're buying

online but you get it if you do dig into

the specifications basically there is a

nine core CPU with three performance

cores and six efficiency cores if you go

for 256 gigs of storage 512 gigs of

storage and 8 gigs of RAM now that's

quite limiting because you're not going

to get you know 16 gigs of RAM and maybe

only 256 maybe you don't want more than

256 you know it's an iPad I know people

use them more like you know laptops and

desktops nowadays but you know this is

quite limiting was if you want the 10

core version you have to go with one

terabyte of storage and get the 16

gigabyt of ram so most people even if

they go with a half a terabyte of

storage and they're just going to have 8

gigs of RAM and the nine core CPU so do

beware when you make your buying Choice

okay so it talked about an improved CPU

micro architecture remember micro

architecture is how the chip is

internally designed each chip M1 M2 M3

the internals are different of course

it's still using it's an Arm based chip

so it's arms architecture but the micro

architecture what's going on the inside

is different in every chip so improved

Branch prediction this is for both calls

Well Branch prediction basically when

you have these pipelines and I've got

videos on this about this here on this

channel you've got these instructions

going down the pipeline they're going

through the decode phase they're getting

along the pipeline getting ready to

execute and then at some point they find

out that what in the pipeline is wrong

because a branch has taken the software

somewhere else and you have to empty the

the pipeline out and then start again

and that causes a performance uh blip

because you've had to empty out the

pipeline and start again so if you can

improve the branch prediction then you

basically you get better performance

fewer Branch mispredictions means

greater performance and that's what

Apple are saying they've done in both of

these uh cord designs now wider decode

and execution engines for the

performance cord well wider pipelines

means more instructions per cycle that

the CPU can fetch decode issue execute

and retire that's the whole kind of

outline of a modern super scaler uh C uh

pipeline uh and so what they're

basically saying is that there's a wider

decode that means more instructions can

be decoded so they come in from the

fetch stage and then they can handle

more of them at the front end uh more

some of them simultaneously and then

towards the back end a wider executional

engine allows for the CPU to execute

more instructions concurrently and that

improves instruction level parallelism

and that's what they've done in this new

processor now they haven't done that in

the efficiency Calles instead with the

efficiency Calles they've got a deeper

execution engine which basically means

there are more pipeline the pipeline is

longer and that probably means there's a

higher clock speed because you can bump

up the clock speed and each clock cycle

now is there's more clock Cycles

happening but you don't have to achieve

so much in each clock cycle because you

split it up into smaller chunks of work

so generally when you make the pipeline

longer you're actually bumping up the

clock speed so I'd imagine that these

new efficiency calls have got a higher

clock speed now here's the key ticket

enhanced NextGen ml accelerators now

these are in the CPU you go back and

look at Apple presentation look at their

diagram these are in the CPU they're not

part of the npu the NP is a separate

thing they talk about the npu separately

talk about its performance talk about

how it's better than previous npus in

the Apple range but this is in the CPU

now why what does it mean now to have an

ml accelerator in the CPU well what

we're talking about is the scalable

Matrix extension specifically the second

version of it which is basically how you

can do matrix multiplication in Hardware

of course we've come a long way from you

know back in the day when a even a

floating Point Unit to do floating Point

numbers was an optional thing and even

today they're optional in

microcontrollers and so on now we're

talking about not even Scala operations

uh but we're talking about Matrix

operations so you can actually do in the

CPU multiplication of matrices and other

Matrix uh kind of operations and that's

good for scientific simulations computer

vision and machine learning and this is

the point now that we've got these ml

accelerators which allow things to

happen in the CPU by allowing these

matrices to be manipulated multiplied

and so on now the key is that2 is part

of the arm

9.4 architecture okay so it's armv 9.4

and it's not available in arm V8 so this

means that this is a new CPU from uh

Apple they've redesigned both cores as

you can see because both cores have got

that uh approved uh uh Branch prediction

both Calles have had their pipelines

change and both cores contain the mlx

accelerators and this is what you do

when you change architecture both sets

of cores have to be the same because

that means that jobs can move from one

to the other and they can actually they

don't say oh no I haven't got the M

accelerator on this on this CPU core and

then it just dies they could the the

jobs can switch from the performance

cause to the efficiency cause cuz they

functionality wise they both do exactly

the same thing so this is Apple's first

arm v9 CPU okay let's dive into the

geekbench six scores I've got both the

single thread scores here and the

multi-threaded scores as you would

expect the single threaded score for

both the nine core version and the 10

core version is better than what you

find in the M3 in this particular case a

MacBook Air I wanted something with

passive cooling the M2 which you found

in the iPad Pro the M1 that you found in

the iPad Pro we can see the progression

here of the single threaded scores and

in terms of multi threaded scores the 10

core version is better than the nine

core version and we can see the 10 and N

core versions are better than the eight

core versions of the M3 the M2 and the

M1 as you'd expect and you can see here

how the M1 the M2 and the M3 have

increased that multi-threaded score

performance as you go up through the

generations however it's important to

note that each of these process is

running at faster and faster clock

speeds so the original M1 was running at

around 3.2 GHz whereas the latest M4 is

running at 4.4 GHz so there's obviously

be more performance when you crank up

the clock speed so if we now actually

divide the single threaded score by the

clock speed we see a very very

interesting thing so trying to eliminate

the clock speed or bring them all down

to the same clock speed what we see is

that for the M1 the M2 and the M3 really

the single threade scores were very very

close 747 743

753 and of course you know geekbench

scores go up and down depending exactly

on the person that run them these are

average scores that I've been looking at

here so really they're not much

different what really happened and I did

say this at the time of the launches of

these of these process what really

happened is they bumped up the clock

speed there were some changes they made

some microarch changes but really they

bumped up the clock speed went down to

different the next nanometer process

node they were able to boost the clock

speed and really if you actually look at

it the M1 the M2 and the M3 are really

very very similar however we can see a

very big leap when we come to the M4 so

even though it's running at 4.4 GHz

which in itself makes things will just

the performance will be greater we can

also see that the actual performance in

terms of the

microarchitecture is much better so

certainly the M4 is a leap uh in terms

of uh Apple's engineering that leaped

arm v9 they've redesigned the CPU they'

redesigned the pipe ples they've

redesigned the branch predictors and

they've up the clock speed because they

now on this second generation 3

nanometer and you've got this great

performance regardless of the clock

speed and and including the new clock

speed and of course we need to compare

this to the Snapdragon X+ and the

Snapdragon X Elite now note neither of

these processes are actually available

in laptops yet we're basing our numbers

on ccoms numbers and of course there

will be some real world testing coming

about when these laptops uh are released

some will have passive cooling some will

have active cooling some will be with a

plus some will be with the elite and so

on and I've covered this before in

previous videos here on this channel but

what do we we see uh M1 this is

multi-threaded score we don't really

have the single threaded score

particularly for the uh uh Snapdragon

X+ so we've got the M1 the M2 and the M3

here with their respective scores then

we've got the M4 with the nine core

version because this is the

multi-threaded one and that is just

beaten by the Snapdragon X+ with its 10

cores now of course as many people will

point out that the nine Core 1 is three

performance cuse and the rest efficiency

CES whereas the Snapdragon X Plus is all

performance caes so that does raise some

interesting questions you've then got

the M4 with a 10 core version and then

next you've got and he's beaten by the

Snapdragon X Elite with the 12 core

version again noting that the M4 with

the 10 core version is with performance

cores and efficiency Calles the

Snapdragon X Elite is with just

performance cores however we don't know

what the power efficiency of the SN

Dragon X Elite process are so if they

are able to pull off this performance

with a similar power enveloper thermal

envelope than the M4 does even though

it's got the two different types of

cores then that makes the Snapdragon ex

Elite an absolute winner if it does it

but it's actually very very power hungry

then that's going to be slightly

different situation however in terms of

overall brute performance in a mobile

laptop type of device then we can see

the Snapdragon X at the moment looks

like like the winner but we're going to

have to see what happens when they

actually come out and see what the

actual power performance numbers are

okay so just moving quickly to the GPU

it's a new 10 core GPU of the M4 Builds

on the existing Graphics architecture of

the M3 so basically this is a tweaked

version of the M3 for the iPad that

means you're getting Dynamic caching now

which is this way of allocating memory

dynamically so you don't just say well

here is you know 2 gigs of memory for

the GPU it goes up and down according to

the gpu's new usage you've got hard

accelerated raay tracing hard

accelerated mesh shading now all these

things we found in the M3 they're now in

the iPad we're going to see them uh

later on in other M4 devices but they've

now come to the iPad so the M4 GPU is

basically an upgrade from the M3 GPU and

this is how uh Apple presented that

during the presentation okay so quickly

overall we've got the M4 built on 3

nanometer from tsmc for you've got 28

billion transistors Hardware accelerated

raid tracing uh av1 uh the 10 core CPU

GPU you can see it all here so a

certainly for the iPad this is amazing

certainly for what we're going to see in

the future predictions well I'm just

guessing I don't have any inside

information I don't have any kind of

leaks to tell I'm just guessing here

there are of course going to be some M4

variance the M4 Pro now could that be a

13 or 14 core with five or six

performance Calles and maybe up to eight

efficiency Calles remember they've added

extra efficiency Calles to the base M4

so that would be interesting M4 Max

could be see even 18 to 20 cores with

again eight efficiency cores we we'll

have to wait and see how Apple stitched

these numbers together how they're going

to do this configuration and we are

going to of course see this CPU in the

iPhone the A8 I would guess now Apple

could leave it for another generation

they could carry on the a18 with what

we've got in the a17 that be the arm v81

I have a sneaking suspicion they're

going to bring this to the a18 so you

get arm v9 in the a18 but remember last

time they called it the A7 Pro so does

that mean we're going to get an a18 Pro

maybe an octacore for the first time

maybe a 3 plus 5 setup I don't think

they've got enough wins in terms of

power to make it a 4 plus4 setup because

those performance Calles are uh Power

hungry because of the performance they

offer so maybe a + 5 that will be

interesting and maybe we can get an A8

without the pro which goes back to the

traditional 6 core 2+4 I don't know I'm

just guessing but Apple clearly have

scope here to do all kinds of

interesting things both in terms of

MacBooks and in terms of the iPhone okay

that's it my name is Gary Sims this is

Gary explains I really hope you enjoyed

this video if you did please do give it

a thumbs up and if you like these kind

of videos why not stick around by

subscribing to the channel okay that's

it I'll see you the next next one

[Music]