10 ms·

The Worst Experience I've Had with an Aarch64 MacBook

58 points by daenney 6y ago

kbar13 6y ago
m1 macbooks seem to be universally loved by developers and are leaps and bounds better than anything previously on the market. i'm so happy for the team at apple, especially given that not so long ago, many of us really really hated the direction apple was going for developers, and they've seem to have not only turned the ship around but strapped on rocket engines
- rvz 6y ago
  If you like M1 you're most certainly going to love an M2 Mac which I would rather wait for. However, I'm interested in the things Apple doesn't tell you. I would also keep an eye out on the internal SSDs since they cannot be replaced [0] especially if you're continuously using intensive software on it or your Mac keeps having to use a swapfile which degrades the flash even further. [0] https://twitter.com/never_released/status/1360657594197671941 https://twitter.com/never_released/status/136065759419767194...
  - nojito 6y ago
    Even at that rate you will be fine for 10 years of use.
    - marcinzm 6y ago
      Yeah, it's more likely that one of the fifty other things soldered to the motherboard will fail before the ssd.
    - pja 6y ago
      Typical guaranteed Tb written limit for a 500Gb 3-bit cell consumer SSD is only 300Tb. (Eg: Samsung 970evo NVMe: https://www.samsung.com/semiconductor/minisite/ssd/product/consumer/970evo/ https://www.samsung.com/semiconductor/minisite/ssd/product/c... ) So if you’re writing 30Tb in 2 months, you’ll start hitting the warranted wear limits in 20 months. I don’t think people expect their Apple laptops to fail in < 2 years. If they’ve gone for (significantly more expensive) 2-bit cell (Samsung 970pro equivalent), then the write limits are 4-5 times greater - Samsung claims 1200Tb for the 970pro. I don’t believe Apple publishes the detailed specifications of their SSD flash chips anywhere though, so it’s impossible to know where the limit is for Apple devices.
      - nojito 6y ago
        Go through the twitter thread. there are other users with much more reasonable numbers. The parent tweet is a clear outlier.
        pja 6y ago
        There are also several with much worse numbers.
      - foldr 6y ago
        It’s been a while now that some MacBook models have had soldered on SSD storage; it’s not new with the M1 Macs. If there were serious reliability issues we’d have seen them already.
niceairport 6y ago
"This website is a fairly complicated webapp written in Rust. As such it makes for a fairly decent compile stress test" From github repo: "My personal/portfolio website." Feels like only one of those two should be true.
- ampdepolymerase 6y ago
  Building CRUD web apps in Rust is fairly difficult as the ORMs and code generators are rather immature and everything must be written out by hand.
- fomine3 6y ago
  It's fine for playground.
- eyelidlessness 6y ago
  Why? Doing cool stuff in your build and making a simple site out of it seems pretty great.
- p1necone 6y ago
  Massively over-engineering personal projects just for the hell of it is a great time, you should try it.
ashton314 6y ago
Does anyone here have experience running Emacs on the new MacBooks? Is the authors experience the norm? My life is an Emacs right now, so this would be a blocker for me to upgrade.
- _-___________-_ 6y ago
  I don’t use emacs often, but I haven’t noticed any hangs with emacs from ports.
- mkskm 6y ago
  I'm running Emacs natively on M1 and it works great. The biggest missing piece is certain static analyzers like shellcheck aren't available for ARM yet since ghc is still working on support, although you can probably get it working with Rosetta. See this thread for a list of most tools: https://github.com/Homebrew/brew/issues/10152 https://github.com/Homebrew/brew/issues/10152 gccemacs isn't available natively yet and probably won't be until later this year, but Emacs running on M1 is very snappy and faster in my experience than a nativecomp build running on an Intel machine.
- amitp 6y ago
  My life is also in Emacs. I'm using it all the time on the new Macbooks and it's working great for me (gui mode, not terminal mode). There are two major versions of Emacs on Mac. There's the "NextStep" version, which is part of the official emacs repository, and the "Mac" version, which is separately maintained. * emacsformacosx.com runs in x86 emulation mode, but flickers when using lsp mode --> this is the version the author used, and I didn't find it to work well. It's the NextStep version. * Mitsuharu Yamamoto's maintains the Mac port and has M1 patches in the "work" branch but not in the master branch. I use the work branch and am very happy with it. Compilation instructions: https://amitp.blogspot.com/2020/11/building-emacs-27-on-apple-arm-m1.html https://amitp.blogspot.com/2020/11/building-emacs-27-on-appl... * in homebrew the emacs and emacs-plus packages both work on arm m1, but I haven't tried these as I prefer the Mac port over the NextStep version. * homebrew "railwaycat" emacs is based on the Mac port, and it looks like it has m1 patches, but I haven't tried it either. * The gccemacs branch was tricky to get working even on my x86 mac, and I haven't tried it on the M1 macbook. It requires libgccjit which is in x86 homebrew but not in arm homebrew (gcc is not yet available). The M1 feels so fast that I've not felt the need to use gccemacs. TL;DR: Emacs is not a blocker, but the widely used version from emacsformacosx.com doesn't work as well as the Mac Port [https://bitbucket.org/mituharu/emacs-mac/src/work/README-mac https://bitbucket.org/mituharu/emacs-mac/src/work/README-mac]
emptyparadise 6y ago
I don't think I want to use Mac OS, but I'd love to have one of these with Linux.
- pid_0 6y ago
  Mac OS is like if linux had a trillion dollar company developing it and an actual integrated ecosystem though You are missing out big time, coming from someone who was staunchly anti-Apple for years
  - new_realist 6y ago
    This. When it comes to desktops, macOS is a national railroad, Linux is a model railroad in someone's basement.
    - emptyparadise 6y ago
      >macOS is a national railroad It's too bad that it's Amtrak and not Japan Railways.
    - wayneftw 6y ago
      Desktops are not railroads, Internet servers are. Considering that the overwhelming majority of Internet servers are running Linux and practically none of them are running macOS, I find this analogy to be hilariously wrong. Even among developers, only ~25% of them run a Mac. Linux has that many developers and the rest is Windows.
      - vineyardmike 6y ago
        >> When it comes to desktops... > Considering that the overwhelming majority of Internet servers... But what about desktops?
        wayneftw 6y ago
        I noted that here: > Even among developers, only ~25% of them run a Mac. Linux has that many developers and the rest is Windows. It's less than 25% among general users...
derefr 6y ago
> The biggest thing that doesn't run properly on here is Emacs. I am able to get a version of it via Rosetta, however there are weird hangs that will randomly eat up all my input while I am in flow. This is undesirable to say the least. This sounds a lot like the kind of stutter Dolphin's dynamic recompiler used to suffer when it first discovered a new shader. IIRC, for faster boot times, Emacs binaries burn in some kind of low-level memory-image that gets dumped into an mmap region and then "revivified." I wonder if that's what Rosetta is choking on. If so, I wonder whether, until an actual AArch64 build is available, it'd be possible to work around this problem by just rebuilding Emacs (for x86!) without that optimization enabled.
- jagger27 6y ago
  Seems like the Homebrew version already builds on M1. People have been working on this since last year. https://amitp.blogspot.com/2020/11/building-emacs-27-on-apple-arm-m1.html?m=1 https://amitp.blogspot.com/2020/11/building-emacs-27-on-appl...
- Iwan-Zotow 6y ago
  > for faster boot times, Emacs binaries burn in some kind of low-level memory-image that gets dumped into an mmap region and then "revivified." unexec hack No true anymore for Emacs 27.x, they go with more portables (though a bit slower) preload
  - derefr 6y ago
    Thank you! I was so confused because I didn't remember the name, and there was no documentation about it I could find without the name (probably because it's out-of-date info now.)
- mkskm 6y ago
  I'm successfully running Emacs natively on M1 using the Yamamoto fork. It just needs to be installed via the formula on Homebrew currently rather than the cask. https://github.com/railwaycat/homebrew-emacsmacport/issues/230 https://github.com/railwaycat/homebrew-emacsmacport/issues/2... Haven't run any benchmarks yet, but general usage is snappier and startup time is around twice as fast as a gccemacs build running on a 2020 Intel MBA.
etaioinshrdlu 6y ago
What's interesting is that they could probably put a higher power (wattage) processor in there and get linearly better multicore performance. But in doing so, the laptop would no longer get great battery life and it would get hot and loud. It benefits Apple to have achieved a level of single core performance that doesn't require high wattage. They may be able to use this to avoid the tendency of applications to eat up any resources available over time and become more power hungry.
Wowfunhappy 6y ago
> The amd64 version of Dolphin uses some Just-In-Time compilation that Rosetta can't emulate at all. That's interesting. Clearly Rosetta can handle JITs since it could run the Intel versions of Chrome and Firefox before native versions of those were released. Are there any other macOS apps which don't work with Rosetta 2? Leaving out stuff like Docker, which is really an entire VM under the hood.
- month13 6y ago
  The dotnet team have been running into some issues getting .NET 5 running well, but improvements are coming with new macOS releases. https://github.com/dotnet/runtime/issues/44897 https://github.com/dotnet/runtime/issues/44897
Teknoman117 6y ago
FWIW, there's not a tremendous amount of parallelism to be extracted here. I pulled the git repo on the site and it compiled in 54 s on a Threadripper 1950X @ 4 GHz (a 16 core processor). The dependencies loaded all of the cores for the first couple of seconds, but most of the time was spent in a lightly threaded section at the end.
hesk 6y ago
Can somebody elaborate on the Emacs issues?
- xena 6y ago
  Article author here. A native emacs build fixes it entirely. I didn't know where a native emacs build was until I got emailed about it after I published the article.
  - hesk 6y ago
    Thank you!
cryptoquick 6y ago
I'm not sure the title and the content of the article quite match, but I'm assuming that's on purpose. I'm looking forward to seeing the M1X / M2!
utf_8x 6y ago
> The biggest thing that doesn't run properly on here is Emacs. Time to switch to a real text editor such as Vim then :) /s
imtringued 6y ago
>This 15 watt laptop chip holds its own with desktop machines Can people stop being surprised that a 15W per core chip is on par with an almost 15W per core chip? Get this into your head. Per core power consumption is not the same as the TDP written on the packaging. Laptops have a high enough TDP that they can run one core at full boost. Desktops have a high enough TDP so that they can run multiple cores at full boost. There is no free lunch there. The ARM Macbook primarily benefits from a better manufacturing process, SoC integration and a better instruction decoder. These factors influence the performance, not some fantasy wishful TDP napkin math. I don't care about Ryzen vs Apple in this scenario. I'm just incredibly annoyed that this ignorance keeps perpetuating itself because people don't understand the difference between per core power consumption and package power consumption and then act surprised that two equal machines (in the context of single core benchmarks) somehow perform the same.
- dangus 6y ago
  > The ARM Macbook primarily benefits from a better manufacturing process, SoC integration and a better instruction decoder It sounds like that just boils down to “it’s a better chip.” You’re even saying that the instruction decoder is better - that means it’s a better chip. What else does a processor do besides decode instructions? How many aspects of the chip do we have to make excuses for before we can just conclude that the M1 is the overall best processor money can buy for a laptop? Consumers don’t care that Apple is using a great TSMC process and Intel isn’t, they care about performance and battery life. The M1 multi-core benchmarks are nothing to scoff at, either. They absolutely keep up with the more expensive, heavier, hotter, worse battery life 16” MacBook Pro Intel models - at least, they keep up close enough to not justify the $1,000+ price premium. When Apple replaces the 16” model, the AS version is probably going to have 16+ cores and make the Intel version look absolutely silly. Just like the Intel MacBook Air, we’ll be asking ourselves how we lived with this x86 trash for so long?
  - hctaw 6y ago
    I think OP is taking issue with the claim that the M1 can compete with desktop class CPUs, which isn't true for multicore benchmarks. The M1 is about on par with current Intel/AMD offerings in single core benchmarks but can't keep up in multicore. Compare: https://browser.geekbench.com/processor-benchmarks/ https://browser.geekbench.com/processor-benchmarks/ to https://browser.geekbench.com/mac-benchmarks/ https://browser.geekbench.com/mac-benchmarks/
xena 6y ago
Author of the post here, ask me anything!
stelcodes 6y ago
Friendly reminder that Apple uses slave labor to build their machines and actively fights against the right to repair movement :)

58 points by daenney 6y ago

m1 macbooks seem to be universally loved by developers and are leaps and bounds better than anything previously on the market. i'm so happy for the team at apple, especially given that not so long ago, many of us really really hated the direction apple was going for developers, and they've seem to have not only turned the ship around but strapped on rocket engines

rvz 6y ago
If you like M1 you're most certainly going to love an M2 Mac which I would rather wait for. However, I'm interested in the things Apple doesn't tell you. I would also keep an eye out on the internal SSDs since they cannot be replaced [0] especially if you're continuously using intensive software on it or your Mac keeps having to use a swapfile which degrades the flash even further. [0] https://twitter.com/never_released/status/1360657594197671941 https://twitter.com/never_released/status/136065759419767194...
- nojito 6y ago
  Even at that rate you will be fine for 10 years of use.
  - marcinzm 6y ago
    Yeah, it's more likely that one of the fifty other things soldered to the motherboard will fail before the ssd.
  - pja 6y ago
    Typical guaranteed Tb written limit for a 500Gb 3-bit cell consumer SSD is only 300Tb. (Eg: Samsung 970evo NVMe: https://www.samsung.com/semiconductor/minisite/ssd/product/consumer/970evo/ https://www.samsung.com/semiconductor/minisite/ssd/product/c... ) So if you’re writing 30Tb in 2 months, you’ll start hitting the warranted wear limits in 20 months. I don’t think people expect their Apple laptops to fail in < 2 years. If they’ve gone for (significantly more expensive) 2-bit cell (Samsung 970pro equivalent), then the write limits are 4-5 times greater - Samsung claims 1200Tb for the 970pro. I don’t believe Apple publishes the detailed specifications of their SSD flash chips anywhere though, so it’s impossible to know where the limit is for Apple devices.
    - nojito 6y ago
      Go through the twitter thread. there are other users with much more reasonable numbers. The parent tweet is a clear outlier.
      - pja 6y ago
        There are also several with much worse numbers.
    - foldr 6y ago
      It’s been a while now that some MacBook models have had soldered on SSD storage; it’s not new with the M1 Macs. If there were serious reliability issues we’d have seen them already.

niceairport 6y ago

"This website is a fairly complicated webapp written in Rust. As such it makes for a fairly decent compile stress test" From github repo: "My personal/portfolio website." Feels like only one of those two should be true.

ampdepolymerase 6y ago
Building CRUD web apps in Rust is fairly difficult as the ORMs and code generators are rather immature and everything must be written out by hand.
fomine3 6y ago
It's fine for playground.
eyelidlessness 6y ago
Why? Doing cool stuff in your build and making a simple site out of it seems pretty great.
p1necone 6y ago
Massively over-engineering personal projects just for the hell of it is a great time, you should try it.

ashton314 6y ago

Does anyone here have experience running Emacs on the new MacBooks? Is the authors experience the norm? My life is an Emacs right now, so this would be a blocker for me to upgrade.

_-___________-_ 6y ago
I don’t use emacs often, but I haven’t noticed any hangs with emacs from ports.
mkskm 6y ago
I'm running Emacs natively on M1 and it works great. The biggest missing piece is certain static analyzers like shellcheck aren't available for ARM yet since ghc is still working on support, although you can probably get it working with Rosetta. See this thread for a list of most tools: https://github.com/Homebrew/brew/issues/10152 https://github.com/Homebrew/brew/issues/10152 gccemacs isn't available natively yet and probably won't be until later this year, but Emacs running on M1 is very snappy and faster in my experience than a nativecomp build running on an Intel machine.
amitp 6y ago
My life is also in Emacs. I'm using it all the time on the new Macbooks and it's working great for me (gui mode, not terminal mode). There are two major versions of Emacs on Mac. There's the "NextStep" version, which is part of the official emacs repository, and the "Mac" version, which is separately maintained. * emacsformacosx.com runs in x86 emulation mode, but flickers when using lsp mode --> this is the version the author used, and I didn't find it to work well. It's the NextStep version. * Mitsuharu Yamamoto's maintains the Mac port and has M1 patches in the "work" branch but not in the master branch. I use the work branch and am very happy with it. Compilation instructions: https://amitp.blogspot.com/2020/11/building-emacs-27-on-apple-arm-m1.html https://amitp.blogspot.com/2020/11/building-emacs-27-on-appl... * in homebrew the emacs and emacs-plus packages both work on arm m1, but I haven't tried these as I prefer the Mac port over the NextStep version. * homebrew "railwaycat" emacs is based on the Mac port, and it looks like it has m1 patches, but I haven't tried it either. * The gccemacs branch was tricky to get working even on my x86 mac, and I haven't tried it on the M1 macbook. It requires libgccjit which is in x86 homebrew but not in arm homebrew (gcc is not yet available). The M1 feels so fast that I've not felt the need to use gccemacs. TL;DR: Emacs is not a blocker, but the widely used version from emacsformacosx.com doesn't work as well as the Mac Port [https://bitbucket.org/mituharu/emacs-mac/src/work/README-mac https://bitbucket.org/mituharu/emacs-mac/src/work/README-mac]

emptyparadise 6y ago

I don't think I want to use Mac OS, but I'd love to have one of these with Linux.

pid_0 6y ago
Mac OS is like if linux had a trillion dollar company developing it and an actual integrated ecosystem though You are missing out big time, coming from someone who was staunchly anti-Apple for years
- new_realist 6y ago
  This. When it comes to desktops, macOS is a national railroad, Linux is a model railroad in someone's basement.
  - emptyparadise 6y ago
    >macOS is a national railroad It's too bad that it's Amtrak and not Japan Railways.
  - wayneftw 6y ago
    Desktops are not railroads, Internet servers are. Considering that the overwhelming majority of Internet servers are running Linux and practically none of them are running macOS, I find this analogy to be hilariously wrong. Even among developers, only ~25% of them run a Mac. Linux has that many developers and the rest is Windows.
    - vineyardmike 6y ago
      >> When it comes to desktops... > Considering that the overwhelming majority of Internet servers... But what about desktops?
      - wayneftw 6y ago
        I noted that here: > Even among developers, only ~25% of them run a Mac. Linux has that many developers and the rest is Windows. It's less than 25% among general users...

derefr 6y ago

> The biggest thing that doesn't run properly on here is Emacs. I am able to get a version of it via Rosetta, however there are weird hangs that will randomly eat up all my input while I am in flow. This is undesirable to say the least. This sounds a lot like the kind of stutter Dolphin's dynamic recompiler used to suffer when it first discovered a new shader. IIRC, for faster boot times, Emacs binaries burn in some kind of low-level memory-image that gets dumped into an mmap region and then "revivified." I wonder if that's what Rosetta is choking on. If so, I wonder whether, until an actual AArch64 build is available, it'd be possible to work around this problem by just rebuilding Emacs (for x86!) without that optimization enabled.

jagger27 6y ago
Seems like the Homebrew version already builds on M1. People have been working on this since last year. https://amitp.blogspot.com/2020/11/building-emacs-27-on-apple-arm-m1.html?m=1 https://amitp.blogspot.com/2020/11/building-emacs-27-on-appl...
Iwan-Zotow 6y ago
> for faster boot times, Emacs binaries burn in some kind of low-level memory-image that gets dumped into an mmap region and then "revivified." unexec hack No true anymore for Emacs 27.x, they go with more portables (though a bit slower) preload
- derefr 6y ago
  Thank you! I was so confused because I didn't remember the name, and there was no documentation about it I could find without the name (probably because it's out-of-date info now.)
mkskm 6y ago
I'm successfully running Emacs natively on M1 using the Yamamoto fork. It just needs to be installed via the formula on Homebrew currently rather than the cask. https://github.com/railwaycat/homebrew-emacsmacport/issues/230 https://github.com/railwaycat/homebrew-emacsmacport/issues/2... Haven't run any benchmarks yet, but general usage is snappier and startup time is around twice as fast as a gccemacs build running on a 2020 Intel MBA.

etaioinshrdlu 6y ago

What's interesting is that they could probably put a higher power (wattage) processor in there and get linearly better multicore performance. But in doing so, the laptop would no longer get great battery life and it would get hot and loud. It benefits Apple to have achieved a level of single core performance that doesn't require high wattage. They may be able to use this to avoid the tendency of applications to eat up any resources available over time and become more power hungry.

Wowfunhappy 6y ago

> The amd64 version of Dolphin uses some Just-In-Time compilation that Rosetta can't emulate at all. That's interesting. Clearly Rosetta can handle JITs since it could run the Intel versions of Chrome and Firefox before native versions of those were released. Are there any other macOS apps which don't work with Rosetta 2? Leaving out stuff like Docker, which is really an entire VM under the hood.

month13 6y ago
The dotnet team have been running into some issues getting .NET 5 running well, but improvements are coming with new macOS releases. https://github.com/dotnet/runtime/issues/44897 https://github.com/dotnet/runtime/issues/44897

Teknoman117 6y ago

FWIW, there's not a tremendous amount of parallelism to be extracted here. I pulled the git repo on the site and it compiled in 54 s on a Threadripper 1950X @ 4 GHz (a 16 core processor). The dependencies loaded all of the cores for the first couple of seconds, but most of the time was spent in a lightly threaded section at the end.

hesk 6y ago

Can somebody elaborate on the Emacs issues?

xena 6y ago
Article author here. A native emacs build fixes it entirely. I didn't know where a native emacs build was until I got emailed about it after I published the article.
- hesk 6y ago
  Thank you!

cryptoquick 6y ago

I'm not sure the title and the content of the article quite match, but I'm assuming that's on purpose. I'm looking forward to seeing the M1X / M2!

utf_8x 6y ago

> The biggest thing that doesn't run properly on here is Emacs. Time to switch to a real text editor such as Vim then :) /s

imtringued 6y ago

>This 15 watt laptop chip holds its own with desktop machines Can people stop being surprised that a 15W per core chip is on par with an almost 15W per core chip? Get this into your head. Per core power consumption is not the same as the TDP written on the packaging. Laptops have a high enough TDP that they can run one core at full boost. Desktops have a high enough TDP so that they can run multiple cores at full boost. There is no free lunch there. The ARM Macbook primarily benefits from a better manufacturing process, SoC integration and a better instruction decoder. These factors influence the performance, not some fantasy wishful TDP napkin math. I don't care about Ryzen vs Apple in this scenario. I'm just incredibly annoyed that this ignorance keeps perpetuating itself because people don't understand the difference between per core power consumption and package power consumption and then act surprised that two equal machines (in the context of single core benchmarks) somehow perform the same.

dangus 6y ago
> The ARM Macbook primarily benefits from a better manufacturing process, SoC integration and a better instruction decoder It sounds like that just boils down to “it’s a better chip.” You’re even saying that the instruction decoder is better - that means it’s a better chip. What else does a processor do besides decode instructions? How many aspects of the chip do we have to make excuses for before we can just conclude that the M1 is the overall best processor money can buy for a laptop? Consumers don’t care that Apple is using a great TSMC process and Intel isn’t, they care about performance and battery life. The M1 multi-core benchmarks are nothing to scoff at, either. They absolutely keep up with the more expensive, heavier, hotter, worse battery life 16” MacBook Pro Intel models - at least, they keep up close enough to not justify the $1,000+ price premium. When Apple replaces the 16” model, the AS version is probably going to have 16+ cores and make the Intel version look absolutely silly. Just like the Intel MacBook Air, we’ll be asking ourselves how we lived with this x86 trash for so long?
- hctaw 6y ago
  I think OP is taking issue with the claim that the M1 can compete with desktop class CPUs, which isn't true for multicore benchmarks. The M1 is about on par with current Intel/AMD offerings in single core benchmarks but can't keep up in multicore. Compare: https://browser.geekbench.com/processor-benchmarks/ https://browser.geekbench.com/processor-benchmarks/ to https://browser.geekbench.com/mac-benchmarks/ https://browser.geekbench.com/mac-benchmarks/

xena 6y ago

Author of the post here, ask me anything!

stelcodes 6y ago

Friendly reminder that Apple uses slave labor to build their machines and actively fights against the right to repair movement :)