Asahi Linux Update Brings Experimental Apple M2 Support

[ddriver]

Yes and llvm provides "ASIC" acceleration for SPEC which helps Apple chips. You won the stupidity award of the month. And that's not the first time you achieved that.

I'd invite you to make an actual case but something tells me you won't.

Yes and llvm provides "ASIC" acceleration for SPEC which helps Apple chips.

Says who? Aside from you obviously.

You won the stupidity award of the month.

Because of what you said?

And that's not the first time you achieved that.

Oh, so you say it has happened before. Therefore logically, it must also be happening right now... Makes sense. Your entire chain of logic checks out!

Seriously, do you even have the simplest idea how basic you are?

[Anux]

of course... sunglasses in the night and "aluminium hat" ....

I was refering to "They Live" a SicFi trash movie from the 80, but worth watching. Even DukeNukem got some lines copyed from that movie.

[Dukenukemx]

No, I'm thanking Apple for designing power efficient hardware, and fixing bugs in their firmware that only affect Linux - see the posts that came before mine.

The hardware you think Apple designed is a mix of ARM and PowerVR technology that Apple merged together. This is evident since the M1 has the same PACMAN bug as other ARM chips do. Also, what makes you think Apple patched their firmware for Linux and not Mac OSX?

[hamishmb]

The hardware you think Apple designed is a mix of ARM and PowerVR technology that Apple merged together. This is evident since the M1 has the same PACMAN bug as other ARM chips do. Also, what makes you think Apple patched their firmware for Linux and not Mac OSX?

I'm not saying that they made an entire ISA from scratch, no, but they did a lot of the work on M1 and M2. Especially eg the Apple Fabric that you find on the M1 Ultra and Max SoCs.

As for Apple fixing things, reading the Asahi blog, though I can't remember which post. Also, a commenter here mentioned this too.

[Dukenukemx]

I think apple deliberately paid for that type of promotion to the tech "journalists" / corporate shills, for a number of reasons, the immediately obvious ones being "look at how glorious our cpu is" and "hide the true origin of improvements at soc level, provide misdirection to the competition". I remember Ian at AT mindlessly parroting their marketing slides, on subjects he has proven time and time again to be conceptually clueless. But the fact is there was no magic or awesomeness to the M1, it was a process node advantage, quite fast integrated memory, and a lot of non-cpu accelerators. This thing had 20 billion transistors, when an 8 core amd apu had about 10. It is evident when it is running general purpose code, and it is evident when it is running translated code that has no way to directly tap into accelerator apis. Any performance or efficiency advantages evaporate, solid proof it is all in the accelerators, and the actual cpu part is ... somewhere in the ballpark of zen 2, just on 5 nm.

I don't see too many people reviewing Apple M series based hardware. The ones that do are primarily Apple only reviewers, meaning they have a bias. Any mention of power efficiency is usually briefly mentioned or extremely limited in testing. Like for example Linus Tech Tips did a battery test... by playing YouTube videos for hours. While this is a great test for a real world use case, it will lean heavily on Apple's Media Engine and not the ARM CPU cores. Also, a lot of people will reference the regular M1 in terms of cooling and power efficiency, but will refer to the M1 Pro, Max, and Ultra for performance. The reality is anything beyond the regular M1 like the Pro and up will use a lot more power and generate a lot more heat. So much so that these M series chips will thermal throttle. Even the M2 based MacBook Air thermal throttles since it doesn't come with a fan, unlike the MacBook 13" with the M2.

The other problem with reviews is what x86 hardware they like to compare it too. Apple only reviewers will of course compare these to older Intel based Macbooks, which will of course lose in every metric. The Linus Tech Tips reviews will use an Intel based desktop CPU with a RTX 3090, because this makes sense when comparing to a laptop SOC like the M series. Of course these setups will eat massive amounts of power. There's no detail on the power settings in the BIOS as most motherboards disable power efficiency settings for better performance. Usually AMD is left out of the benchmarks and if an AMD is used then it's a Threadripper. When in reality you should be comparing these to a mobile SoC like the 6900HS. They do make desktop PC's that use these chips, like what Apple did with their desktop M series variants.

This is by far the most comprehensive review of a Apple M series based product. Goes into great detail about power consumption.

What I'm disturbed is the amount of Apple apologists here on this Linux forum. Apple is one of the worst when it comes to open source support, and even worse when it comes to Linux support. Apple is no friend to Linux. Apple letting you install Linux is not them being pro Linux. Have we forgotten the T2 ‘secure enclave’ chip that prevents you from installing Linux? Pro Linux means they open sourced their drivers for all the functionality of their M series chips. You know, like how AMD and Intel have done for over a decade. Apple dumped CUPS in favor of IPP, you know the printing software we Linux users use. Mac OSX isn't so different from Linux that Apple couldn't just release it. You cannot seriously be pro Linux and Pro Apple at the same time, and do it with a straight face.

[hamishmb]

Here's a good one on why people compare Apple Silicon to powerful desktop systems:

https://www.youtube.com/watch?v=-XO3...ist=LL&index=5

TL;DW: For video editing, the base Mac Studio beats a Threadripper with RTX 2000 series GPU for video rendering and tracking with Da Vinci Resolve. Unlike others, this is using cross-platform software rather than just comparing eg Final Cut results - it's more fair.

This is why we compare it to a Threadripper - because it's better in some ways. Of note, Linux also uses IPP. And Apple isn't the one dominating the market and disabling (by default) 3rd party signing certificates for secure boot. I'm the first one to talk about not liking Apple or their ethics, but they do a lot of things right.

[Dukenukemx]

Here's a good one on why people compare Apple Silicon to powerful desktop systems:

https://www.youtube.com/watch?v=-XO3...ist=LL&index=5

TL;DW: For video editing, the base Mac Studio beats a Threadripper with RTX 2000 series GPU for video rendering and tracking with Da Vinci Resolve. Unlike others, this is using cross-platform software rather than just comparing eg Final Cut results - it's more fair.

This is why we compare it to a Threadripper - because it's better in some ways.

Video editing on the M series is better because of fixed function hardware. Apple spent a lot of money for this purpose. It's so powerful that the video rendering on the M1's is limited by the SSD. And yes Da Vinci Resolve does support Apple's Media Engine. This isn't using the ARM CPU's or even Apple's GPU, but the Media Engine hardware. To give you an idea of how massive that Media Engine must be, the M1 Max found on the Mac Studio that reviewer was using is 57 billion transistors. Compared to the 32 core Thread-ripper which uses 19 Billion. The RTX 2080 Ti he's using is 19.6 Billion. Either case, these are more general purpose hardware, though I would argue that the RTX 2080 Ti is made for gaming not video rendering.

People here do understand that fixed function hardware is only good at one task and one task only. it can't do other tasks.

Of note, Linux also uses IPP. And Apple isn't the one dominating the market and disabling (by default) 3rd party signing certificates for secure boot. I'm the first one to talk about not liking Apple or their ethics, but they do a lot of things right.

What exactly does Apple do right? I honestly can't think of a one. Apple doesn't allow others to make clones of their products. You can't repair their products. I can replace the SSD in most Windows based laptops, where as I can't with any of Apple's M series laptops. You do realize that the M series products have a high read write rate on their SSD's? Those things will eventually wear out and can't be replaced. if it isn't SSD wear then it's the random event where sometimes 13 volts is sent straight to the SSD, and thus killing it with no hope for repair, let alone getting your data.

As far as Linux goes, Apple is the exact opposite of what the linux philosophy is about. They Deprecate cross-platform standards to push their own ABSOLUTELY. PROPRIETARY standards. At least I can run OpenGL and Vulkan on Windows. Good OpenGL, as Apple left it behind years ago. You cannot claim Apple is good because Microsoft. They're both bad, but Apple is a different kind of bad. Mac OSX isn't even open source for crying out loud.

https://youtube.com/clip/UgkxfK-t2EiJDnZTNQ_2dl42Epi3V1HTkA1o

[hamishmb]

Yes, they do a lot of stuff wrong too.

The high read-write rate was caused by a software issue and was fixed IIRC. Also, they provide spare parts for some of their newer phones. They also need to provide spares for their older phones, but progress is progress.

There are also a lot of reasons why I dislike Apple and macOS and iOS, I hasten to add, but I can see many positives in there too.

Just because the hardware is fixed function doesn't mean the performance gain isn't there, either. If it does what people want it to do, does it well, and does it at a good price, then Apple have designed their chips well for their customer base, in my opinion.

EDIT: Also going to note that like any piece of silicon ever, there are going to be things that it's not so good at. That's fine, and quite normal. If you eg need to run Windows, or Windows specific software, this isn't for you, obviously. Especially as Wine is still not quite there, though I hear crossover does well on Apple Silicon. You can also run Vulkan with MoltenVK, and OpenGL on top of Zink on MoltenVK. I know it isnt perfect, and Apple shouldn't have ditched OpenGL, but every platform has its problems.

Anyway, this is getting more flamey and further away from facts, so I'll exit the conversation here - this isn't contributing anything useful unless there's some new information or just general discussion.

[BHZeto]

It's Linux on the M1/M2 Macs. The use case is whatever you want to make of it.

There are lots of way cheaper ARM devices that readily support Linux. "Whatever you want" is not an answer, tell me what's the use case.

[qarium]

Most of the performance miracles on the M chips is due to asic acceleration, not general purpose compute as some perpetuators of uarch myths of awesomeness would have people believe. Running translated code, or running code that doesn't benefit by those accelerators reveals the actual performance characteristics of the cpu itself, rather than the "soc performance".

I always say it is very important to have benchmarks for both. Today we largely have "platform" performance tests, with many optimized routines running on different purpose specific hardware, which is just not a good indication of how well the actual cpu component is capable of on its own.

I think apple deliberately paid for that type of promotion to the tech "journalists" / corporate shills, for a number of reasons, the immediately obvious ones being "look at how glorious our cpu is" and "hide the true origin of improvements at soc level, provide misdirection to the competition". I remember Ian at AT mindlessly parroting their marketing slides, on subjects he has proven time and time again to be conceptually clueless. But the fact is there was no magic or awesomeness to the M1, it was a process node advantage, quite fast integrated memory, and a lot of non-cpu accelerators. This thing had 20 billion transistors, when an 8 core amd apu had about 10. It is evident when it is running general purpose code, and it is evident when it is running translated code that has no way to directly tap into accelerator apis. Any performance or efficiency advantages evaporate, solid proof it is all in the accelerators, and the actual cpu part is ... somewhere in the ballpark of zen 2, just on 5 nm.

x86 is getting there currently, but since it is n hardware vendors running a number of different os and software from scores of vendors, it will take longer to properly leverage that. Apple in comparison controls everything top to bottom, they have the guarantee what hardware baseline all software will run on, they can integrate asic functionality on system library level, into the app sdk, and the benefits are automatically propagated to 3rd party software with little to no need to manually optimize anything.

thats all right but please watch:
" The future of computing: a conversation with John Hennessy (Google I/O '18)"

https://youtu.be/Azt8Nc-mtKM

it is impossible to do: "general purpose compute" fast it is plain and simple impossible.

the only way to do it fast with low power is by "different purpose specific hardware" means "integrate asic"

and this has nothing to do with ARM VS x86_64 the ARM vs x86 part is only about 5% of the tranistors on the general purpose compute part in the cpu...

So really do not know what is your problem the John Hennessy video clearly shows that put in more "different purpose specific hardware"+ "integrate asic" party make it fast.

in general apple M2 is just better better nm node better cpu ISA(more simple to unterstand in assembler and 5% less tranistors compared to x86) and what also makes it better more ASIC parts and so one and so one.

and as you said the Wintel monopole (microsoft/intel) is just slower in adopting this "more ASIC is faster" road.

we as linux users why should we buy intel hardware if the apple hardware is better ? only because intel works on linux drivers ? apple could already do it to in secret make an NDA with an developer pay him but make im do so silencly .. we do not know.