Apple M Chips - The End. Was it even worth it?

In 2020, Apple revolutionized the world of computers with its M1 chip. In this video, we'll cover all the advancements and challenges the company has faced on its journey to perfecting its chips. Find out how competitors caught up to Apple in power and energy efficiency, and why the switch to a new architecture in the M3 chip didn't yield significant performance gains. We'll also discuss Apple's future plans for artificial intelligence and optimizing its devices, as well as tips for Mac buyers.

00:00 Apple M Chips.
00:23 Do you remember the M1?
02:03 The problem with M processors
04:13 Apple's Competitors
05:10 All in one chip
06:04 Apple's biggest problem
06:44 Apple's 3 solutions
08:44 Real changes
10:28 The problems continue...
12:17 The M3 wasn't supposed to show up?
14:05 New production for Apple
16:25 Apple's AI
18:25 Should we buy a Mac now?

-----------------------

GEAR THAT I USE ↓

As an Amazon Associate I earn from qualifying purchases.
-----------------------

-----------------------

[QuantumCanvas]

When you ask GPT to write scripts for you video

[TheRockingest]

I was all fired up to bash this video, but after reading the comments, it looks like everything has been addressed! I have faith in humanity!

[jameshewitt]

"Without getting too technical" - proceeds to demonstrate that the reason you aren't getting too technical is because you literally don't understand it on a technical level.

[iokwong]

Yet, another YouTuber who has no idea what they are talking about when it comes to CPU instruction

[MarbsMusic]

Tell us you don't understand processor design without telling us you don't understand processor design...

[Holden_McHock]

Bro's getting destroyed in the comments 💀

[michaelashby]

30% gain isn't impressive?! Ok, let's see you improve the performance of anything in computer hardware or software by just 1%.

[beragis]

Arthur did not do research, at the 10:36 mark he says that TSMC can not go lower than 3nm. They mentioned in April moving to 1.6nm which is smaller than Intel's 1.8nm. Samsung, Intel and TSMC are all racing to the 1nm barrier with a goal of 2030.

[amritrosell]

If you instead of making assumptions about architecture learn a bit deeper about the differences between the various nodes of 3nm architecture that the chips are made on you would perhaps realize M3 was more of marketing strategy from Apple to be the first CPU that used 3nm, but they did it on the same "dead" branch of 3nm node because the branch the M4 uses is very different from M1-M3 uses. The reason why there was so little improvement is mainly because they didn't redesign the chip particularly much except from removing some parts that M2 Ultra uses to use that area for other things. But now with M4, the architecture is using a very different 3nm node AND they have done an overhaul of the chip design as we can see on the M4 in iPad.

So yes, M3 was a bit of a shrude move and a whole lot of shenanigan and mostly marketing just to be the first CPU that uses 3nm.

But to use the numbers from M2 to M3 and get the M4 numbers is probably not going to be so correct as both the new 3nm node is much more efficient and the chip design is overhauled to accommodate the efficiency of the new node... So, no, Apples chip design isn't dying, it's evolving. But sometimes they jump onto things just to be first snd that might look Peculiar...

[llampp]

2:14 you're saying that a 15% speed increase per generation isn't "that much"? THAT'S MASSIVE for a single generation jump.

[vernearase]

If you don't understand computer architecture and processor design, just make shit up based on what you _think_ is going on.

Just adding transistors don't make the processor faster - adding decoders and making the pipeline deeper with a lot of instruction prefetch and execution reordering and branch prediction make things faster; the transistor count increases to support these things.

Nuvia was formed from ex-Apple silicon engineers, and the Snapdragon X was a joint project of ARM and Nuvia engineers. They collaborated to build a server chip, then Nuvia was acquired by Qualcomm - so the Snapdragon X is pretty much a bastard child of Apple.

The M4 is built on TSMC's N3E node whereas the M3 was built on TSMC's N3B node - a custom node built for Apple when N3E wasn't going to be ready in time (and Apple _really_ wanted a 3nm processor). N3B is more complicated to manufacture and has lower yields, whereas N3E is on TSMC's official 3nm roadmap and is compatible with future nodes like N3P which will result in lower cost. M4 has higher memory bandwidth and is (I believe) Apple's first ARMv9 chip.

[BeaglefreilaufKalkar]

The famous German philospher Nuhr once stated: "If you don't have a clue, just shut the f*ck up."

[swdev]

Is this the spiritual successor to The Verge PC build video?

[diebygaming]

You never mentioned the actual reason they don't just make arbitrarily large chips is because increasing the size of the chip decreases the yield

[SpaceTimeAnomaly]

The 5nm or 3nm is NOT the transistor size -- it is only the size of the smallest structure.

[hyksos]

Arthur, you really need some help to do 'tech' proofing, many thing you said is wrong:
- ARM vs x86: modern X86 processor no longer have dedicated circuitry for each instruction, they decode it into simpler instructions, btw please prononce it Arm like a single word. And the reason why Arm is superior to X86 is mostly around the way it handle its memory access.
- Process density: TSMC N5 N3 N2 etc... are marketing, they do not represent the real transistor size and can't be use to deduce transistor density, but between each major number reduction there's a technology leap, using new type of gate, adding an extra patterning mask etc... and the reason why the M3 was created on N3 process is that Apple did a deal with TSMC to take 100% of the process production, TSMC was struggling with this process and wanted to reduce the number of client on it to optimize the process for that single client, Apple took the deal because it was still giving better result then N5 process and is good for marketing.
- AI TOPS: it is not the data set that is different size, it is each of the data element that is, this is a HW implementation and is why the previous generation M chip can't double its performance using smaller data size. Btw not all AI task can use that small data size, it is still pertinent to compare FP16 performance, simply divide the XElite and M4 number by 2. But for AI task that can work with in8, the XElite really is 2.5X better then M3 (and 4X the M1) I really don't understand why Apple did not put the A17's NPU on the M3, my guess is that the M3 chip was created before the A17 (even if it was released after) to run production test on N3 process, to de-risk A17 production.

I know that 99% of your listener won't notice any of these errors, but it can irritate many technical peoples :)

[Suqrat]

He doesn't know anything about chip design or manufacturing. Very bad info in this video. I am Electronics Engineer with 15 years experience in chip design field.

[Noobtaco]

Who calls it A.R.M? No one. It’s arm. 💪

[khyleebrahh]

Bring back dislike views. To stop miss information

[BrentLeVasseur]

I’m watching this video on my new M4 iPad Pro. It plays Youtube superfast! I was able to watch this video in half the time!😂