Apple’s M1 chip fastest laptop CPU in the world

[Klaus]

Hm, interesting, but I wonder why they don't simply recompile Windows and its applications with an ARM compiler and create native ARM binaries? Why is binary compatibility so important? Should be simple to recompile Windows and the biggest standard applications (Office, Adobe stuff, ...) and then live with slower performance for those applications that aren't available as native ARM binaries.

[Axtremus]

FireFox version 84 with native Apple M1 silicon support is out.
Yes, it feels faster/snappier than FireFox on Intel Core i5.

[Axtremus]

Got my simple performance comparison between my Intel-based MacBook Air (2018, 1.6 GHz Dual-Core Intel Core i5) and my Apple M1-based MacBook Air (2020).

Got both systems to do exactly the same task and timed them using the Unix "time" function, I got:

• Core i5: 575.74 real. 571.99 user. 0.91 sys
• Apple M1: 202.37 real. 202.00 user. 0.41 sys

Think of it as a single thread, single core comparison. I am fairly certain my code's core loop is small enough to fit entirely in cache, the problem does not lend itself to parallelism, and there is barely any I/O, so this is a CPU-centric comparison. Doing exactly the same task, the Apple M1 CPU takes only 35% of the time taken by the Intel Core i5 CPU. Or put another way, a single core of the Apple M1 CPU is roughly 2.8 times faster than a single core of the Intel Core i5 CPU.

Back when playing with the circle of light bulbs puzzle, I wrote a simple program to simulate that puzzle. That's what I used to compare the Intel Core i5 MacBook Air and the Apple M1 MacBook Air -- the comparison was to simulate that puzzle for a ring of 36 light bulbs.

The reason I waited this long to do this comparison is to wait for Go lang to support the M1 silicon natively, and that support was finally released on 2021-02-16.

The M1 is indeed very efficient. Running that simulation of 36 bulbs, the Intel system got really hot to the touch and I could hear the fan kicking into high gear. The M1 system stayed cool to the touch through out and it does not even have a fan.

[Axtremus]

Adding one more comparison:

• Core i5 (2018 MacBook Air) : 575.74 real. 571.99 user. 0.91 sys
• Xeon E5 (2013 Mac Pro) : 507.49 real. 507.43 user. 0.41 sys
• Apple M1 (2020 MacBook Air) : 202.37 real. 202.00 user. 0.41 sys

[Axtremus]

Another comparison, these are for a floating-point heavy computation task, exponentiating a close-to-one floating point number to some ridiculously high degree :

• 1.7 GHz Intel Core i7 (2013 MacBook Air): 206.06s
• Apple M1 in Rosetta 2 emulation (2020 MacBook Air) : 116.02s

[Klaus]

I encountered a very typical M1 compatibility issue today: I have a Canon camera that can be used as a webcam when installing the "EOS Webcam Utility", but it turns out that it doesn't work on the M1. Most of the big software packages run without major problems, but it's always the little things that are problematic. All in all, I should have bought an Intel Mac. The performance benefits are rarely significant in practice, but the sum of all the minor and major compatibility issues are.

[Axtremus]

@axtremus said in Apple’s M1 chip fastest laptop CPU in the world:

Even more comparisons (new ones added 2021-08-03, rearranged from fastest to slowest):

• Apple M1 (2020 MacBook Air) : 202.37 real. 202.00 user. 0.41 sys
• Xeon E5 (2013 Mac Pro) : 507.49 real. 507.43 user. 0.41 sys
• Core i5 (2018 MacBook Air) : 575.74 real. 571.99 user. 0.91 sys
• netcup "VPS 1000 G9": 580.35 real. 568.84 user. 13.05 sys
• aws "t2.small": 713.58 real. 702.96 user. 0.10 sys
• aws "t2.micro": 744.17 real. 743.85 user. 0.00 sys

The "netcup VPS 1000 G9" refers to the same type of server used to host TNCR, from these results, you can see that its "compute" performance is roughly the same an Intel Core i5 used in a 2018 MacBook Air. (Klaus, don't worry, I ran the test on a separate server, I am now also a netcup customer and took out a "VPS 1000 G9" separately myself. )

The "aws" items refer to Amazon Web Services' two different server platforms. The "aws t2.small" costs more than twice the "netcup VPS 1000 G9" but perform worse anyway, but AWS allows me to very flexibly rent it "by the hour". netcup provides much lower prices but the low price requires a one-year commitment.

The "t2.micro" is just an even slower, even cheaper server option from AWS. It's a "teaser" item provided by AWS to entice hobbyists and perhaps events students/academics to try out AWS, in that AWS will let you take out a "t2.micro" for "free" with various limitations.

With these comparisons, I am quite impressed that netcup can provide a fairly high-performing server (with tons of storage/bandwidth) this economically, and that Klaus managed to find it when looking for a new home for TNCR.

[Klaus]

@axtremus said in Apple’s M1 chip fastest laptop CPU in the world:

With these comparisons, I am quite impressed that netcup can provide a fairly high-performing server (with tons of storage/bandwidth) this economically, and that Klaus managed to find it when looking for a new home for TNCR.

To be honest, the main reason why I stumbled upon Netcup was a Google search for "2 girls 1 cup".

[George K]

Link to video

[Copper]

Speed isn't everything
.
.
.
.
.
.
.
.
Just kidding

[Klaus]

Well, the problem these days is: speed isn't speed. A CPU that is fast at doing one thing may be slow at doing another. It also depends a lot on how well the programmers and compiler writers optimize their programs for a particular kind of CPU.