1
00:00:00,000 --> 00:00:04,680
A short while ago, we did a video on what we might expect from Apple's in-house silicon

2
00:00:04,680 --> 00:00:09,040
after it was announced this past summer, but now, the first systems with the Apple M1,

3
00:00:09,040 --> 00:00:12,840
the inaugural chip in the new Apple Silicon series, are on the market.

4
00:00:12,840 --> 00:00:17,200
So, let's take a closer look at how the chips perform and the innovations under the hood

5
00:00:17,200 --> 00:00:20,280
that Apple hopes will make the M1 a home run.

6
00:00:20,280 --> 00:00:25,040
A cursory glance might make you think that the M1 is just a souped-up mobile chip as

7
00:00:25,040 --> 00:00:30,320
it is quite similar to the A14 in the current iPhone 12 lineup, with four high-performance

8
00:00:30,320 --> 00:00:37,240
and four higher-efficiency cores. And it has IO controllers and built-in memory that makes it more of a mobile system on a

9
00:00:37,240 --> 00:00:44,840
chip than a plain CPU. But Apple claims that it's the world's fastest laptop CPU, and early reviews actually show

10
00:00:44,840 --> 00:00:51,440
that this claim isn't really unfair, with many commentators gushing over its performance.

11
00:00:51,440 --> 00:00:56,760
But how did Apple pull this off without fundamentally changing the design from what it's using in

12
00:00:56,760 --> 00:01:04,560
its smartphones? Part of the answer is that Apple used a lot more transistors in the M1 than the A14.

13
00:01:04,560 --> 00:01:12,600
Over four billion more, according to the company. But while this may have boosted performance, the real secret sauce lies in how Apple's

14
00:01:12,600 --> 00:01:19,280
design, both in the A14 and the M1, fundamentally differs from x86, which is the architecture

15
00:01:19,280 --> 00:01:27,200
that AMD and Intel have been using for decades. Apple's philosophy was to make the chip much more parallel than conventional CPU designs.

16
00:01:27,200 --> 00:01:32,400
Both the M1's decoder, which translates incoming instructions, and the execution units, that

17
00:01:32,400 --> 00:01:38,200
actually process them, are wider, meaning they can accept more instructions at once.

18
00:01:38,200 --> 00:01:43,680
Additionally, the M1 can go significantly deeper with out-of-order execution, meaning

19
00:01:43,680 --> 00:01:48,720
that it can read ahead on the page and anticipate which instructions a program will need to

20
00:01:48,720 --> 00:01:54,160
have processed ahead of time to a greater extent to that of x86.

21
00:01:54,160 --> 00:02:00,960
Then you have the fact that the M1 features a lot more level 1 cache than x86 processors,

22
00:02:00,960 --> 00:02:04,320
which is the fastest cache memory available to a CPU core.

23
00:02:04,320 --> 00:02:09,400
And because it's based on a chip originally meant for mobiles, it does all this while

24
00:02:09,400 --> 00:02:17,440
drawing significantly less power. But hold on a second, couldn't Intel and AMD just implement some of the changes themselves

25
00:02:17,440 --> 00:02:24,280
and catch up with Apple? Well it turns out that might be quite a challenge, as the x86 architecture has some inherent

26
00:02:24,280 --> 00:02:30,840
limitations. For example, AMD and Intel could just try and add more L1 cache, but it's extremely

27
00:02:30,840 --> 00:02:37,000
difficult to make decoders much wider than they are now on x86 chips, so can't win

28
00:02:37,000 --> 00:02:42,680
them all. So does this mean that Intel and AMD are in huge trouble?

29
00:02:42,680 --> 00:02:47,360
Well, like any new impressive piece of tech, the M1 isn't without its weaknesses.

30
00:02:47,360 --> 00:02:51,880
As we see, although early reviews have reported very impressive performance gains in first

31
00:02:51,880 --> 00:02:57,040
party apps that have been written for Apple Silicon, the M1's speed benefits in other

32
00:02:57,040 --> 00:03:03,160
applications have been more modest, as programs written for x86 have to be emulated, which

33
00:03:03,160 --> 00:03:06,240
introduces significant CPU overhead.

34
00:03:06,240 --> 00:03:10,960
And because the emulation process is imperfect, there have also been stability issues with

35
00:03:10,960 --> 00:03:16,020
a number of popular programs, even though they were written for late model Mac hardware.

36
00:03:16,020 --> 00:03:20,620
But Apple is betting that because developers who have large user bases on the Mac platform

37
00:03:20,620 --> 00:03:26,020
would like to keep it that way, those developers will adapt and come out with versions specifically

38
00:03:26,020 --> 00:03:32,900
written for the M1 sooner than later. This approach is more or less in line with Apple's playbook for other new products.

39
00:03:32,900 --> 00:03:37,180
They bring a new platform to market and then use their brand power to force developers

40
00:03:37,180 --> 00:03:42,340
to catch up, which is something that they hope will again happen with the M1.

41
00:03:42,340 --> 00:03:47,220
Ultimately, it's not hard to imagine offering a more tightly controlled App Store experience

42
00:03:47,220 --> 00:03:52,340
for Macs, with programs being specifically vetted for full compatibility with Apple Silicon,

43
00:03:52,340 --> 00:03:56,620
especially as we expect to see more of Apple's chips in desktop Macs down the line after

44
00:03:56,620 --> 00:04:01,340
seeing how powerful their first crack at a laptop processor has been.

45
00:04:01,340 --> 00:04:04,980
Hmm, not even the mood for some Apple chips for the side of peanut butter.

46
00:04:12,340 --> 00:04:13,900
Thanks for watching, and I'll see you in the next video.