HN Explorer

Ask HN: What technically allows the M1 to perform better?

5 points by light_hue_1 ↗ HN
I'm flabbergasted that Intel and AMD have failed so utterly and completely in every respect. These are huge companies with half a century of lead time and they're now years behind Apple's M1.

There's a lot of speculation about why the M1 is faster. What's the actual answer? It can't just be the ISA, the M1 performs superbly when running x86-64 code too. What does the M1 do chip-design-wise that Intel and AMD failed to do for their chips?

10 comments

[ 2.5 ms ] story [ 36.8 ms ] thread
majormunky
I think part of it is having chip people being able to work closely with the people writing software for it. I remember something about the M1 chip being able to do simple things like releasing an object and it taking much less time than on x86 as an example. https://twitter.com/catfish_man/status/1326238434235568128?s...
oumua_don17
You are right.

SJ quotes Alan Kay https://youtu.be/XAfTXYa36f4

This was some 15 years back and it seems it was always Apple long term vision to be where they are today with M1. It would be completely fascinating to see how such vision was realised across two CEOs and multiple C level changes combined with a secretive culture.

Any ex-Appleites here to throw some light?

getcrunk
The m1 had the "luxury" of being able to be made from the ground up relatively speaking using modern knowledge.

If you consider the case of Intel/and and x86 imagine a building with one story and every year they had to add a story to it. That iterative improvement added some cruft and they also didn't have the option to rebuild from scratch.

Apple has great minds working on the m1 and alot of resources and I'm sure they've made unique accomplishments but the biggest thing is that I think.

wmf
Apple has been iterating on their cores for around seven years, AMD's Zen is around four years old, and Intel redesigns their cores occasionally. The ARMv8 ISA is less crufty than x86-64 though.
getcrunk
That's a good point but my assumption is apple planned the m1 from around the same time as their ax core
ThePhysicist
Most of it is due to Apple using a more modern fabrication process with smaller structures (5nm). Some of it is due to optimizations like having efficiency cores that can handle non-demanding workloads with less energy consumption than the regular cores. I expect Intel & AMD chips to quickly catch up once they switch to more modern processes as well. Currently Intel still uses a 14 nm process for their Coffee Lake processors, so no matter how good their design might be they just can't compete with a chip that's built using a 5nm process, as the physical characteristics are completely different.
gnatman
School me- I always thought those process numbers were more marketing metrics than directly comparable measurements?
ThePhysicist
In a nutshell, smaller transistors draw less energy at a given operating frequency. According to TMSC the 5 nm transistors draw 30 % less power as compared to their 7 nm ones [1]. Of course the devil's in the details, but a large fraction of the performance and efficiency gains can be attributed to the process.

[1]: https://en.wikipedia.org/wiki/5_nm_process