Apple Silicon M1: Black Magic Fuckery • Hacker News Board

Re: Apple Silicon M1: Black Magic Fuckery

#611

Post by dirtypersian » Wed, Nov 25, 2020, 9:04 AM UTC

dannyw wrote: post #540

dirtypersian wrote: post #515

Earlier quoted context omitted.

so you're saying somehow Rosetta2 is looking at an x86 binary and figuring out exactly which portions of the program rely on the TSO ordering for correctness and then dynamically switches to weak ordering for parts that might be able to do without? I don't really know much about the internals of macOS but figuring out when there are applications for example running on two different cores (since TSO is only really nee…

No; it’s not explained very well but the M1 chip features hardware support for TSO which Rosetta2 uses. It’s really ‘Apple Silicon’ and not just ARM.

> It's really 'Apple Silicon' and not just ARM.

Yeah, I think that's key to understanding this. They are supporting a version of ARM ISA running that maintains TSO even though official ARM doesn't need to support TSO. I guess this is all to get better emulation performance and avoid those extra synchronization instructions that would have to be added by Rosetta if the silicon did not have TSO support.

Re: Apple Silicon M1: Black Magic Fuckery

#612

Post by izacus » Wed, Nov 25, 2020, 9:05 AM UTC

ogre_codes wrote: post #553

mathnmusic wrote: post #539

Earlier quoted context omitted.

The parent was complaining about being forced to use OS X. So, the "walled garden" is intended for the underlying hardware.

> So, the "walled garden" is intended for the underlying hardware. I don't think that is what he was saying at all. Regardless. There is nothing preventing Linux (or Windows) from booting on Mac M1 hardware. People almost certainly will have Linux running on Mac hardware before too long. It's just a slog getting it working well. Not a walled garden in any traditional sense, just difficult to implement.

> There is nothing preventing Linux (or Windows) from booting on Mac M1 hardware.

Except the iOS-like secure boot verification and complete lack of any kind of documentation or drivers for the hardware. Details!

Re: Apple Silicon M1: Black Magic Fuckery

#613

Post by vermilingua » Wed, Nov 25, 2020, 9:07 AM UTC

For those of us who are picking up M1 MacBooks as first time Mac users, is there some kind of Mac crash course for devs? What apps are useful, what familiar tools from Linux can we use, etc? I'm aware of Dash, which makes me suspect there are a bunch of other Mac-exclusive tools which will be useful.

Re: Apple Silicon M1: Black Magic Fuckery

#614

Post by Hamuko » Wed, Nov 25, 2020, 9:07 AM UTC

foldr wrote: post #602

nicolas_t wrote: post #589

Earlier quoted context omitted.

Exactly this, I was fine with the Mac Pro until 2012, it was a little more expensive than PCs but not much outrageously so (maybe 30% more, that's a tax I was fine paying given the OS and that it was quite a well built machine). The new Mac Pro is 3-4x the price of a machine built around AMD having equivalent performances. I'm building a Threadripper for exactly this reason. Most of the issue is Intel vs AMD and the…

There is also the iMac and iMac Pro range. The Mac Pro is clearly demarcated as a "money is no object" product.

I'd be much more interested in the iMac if it wasn't built inside a monitor.

Re: Apple Silicon M1: Black Magic Fuckery

#615

Post by amelius » Wed, Nov 25, 2020, 9:11 AM UTC

favorited wrote: post #454

brundolf wrote: post #54

Earlier quoted context omitted.

I'm wondering if the "optimized for reference-counting" thing applies to other languages too. i.e. if I write a piece of software in Rust, and I make use of Rc , will Macs be extra tolerant of that overhead? In theory it seems like the answer should be yes

No, because Rc isn't atomic. Arc , however, would get the benefit. The reason retain/release in ObjC/Swift are so much faster here is because they are atomic operations.

So, useful for structural sharing between threads.

Re: Apple Silicon M1: Black Magic Fuckery

#616

Post by quantummkv » Wed, Nov 25, 2020, 9:12 AM UTC

sudosysgen wrote: post #532

quantummkv wrote: post #502

Earlier quoted context omitted.

Given that Linux desktop cannot properly utilize even the limited amounts of very common HW accelerators build into AMD and Intel chips (video decoders) and the existing GPU's (all of them suck in different ways), running Linux on M1 would be a waste of everything without giving Linux any noticeable improvement over existing x86 machines.

Accelerated video decode worked fine on my AMD card, actually. Some codecs weren't supported, but it was pretty much functional. On Nvidia, I literally just had to install one package and I never had to think about video decode acceleration again.

It still doesn't work in browsers correctly, which is the most common way to consume video on Linux desktops

Re: Apple Silicon M1: Black Magic Fuckery

#617

Post by skohan » Wed, Nov 25, 2020, 9:12 AM UTC

oblio wrote: post #582

skohan wrote: post #577

Earlier quoted context omitted.

Isn't this mostly about driver support? I'm not a hardware guy, but it's hard to understand why it's so difficult to reverse engineer something like the touchpad driver. Isn't it just a matter of measuring the IO and reproducing it? I mean how has MS done this for bootcamp?

Well, it doesn't seem to happen in practice so it's either hard or there are no incentives to do it, or both. But that's the reality we live in.

Yeah fine, you're not wrong, but it's still worthwhile to understand why it's the reality we live in verses just taking a defeatist attitude.

If it's just a matter of one engineer taking a couple weekends to change the situation, that's surely worth interrogating.

Re: Apple Silicon M1: Black Magic Fuckery

#618

Post by indemnity » Wed, Nov 25, 2020, 9:16 AM UTC

lazyjones wrote: post #571

yurishimo wrote: post #434

Earlier quoted context omitted.

Tbf, WoW will run on just about anything... I remember playing it on an Acer (or maybe Lenovo?) netbook back in 2009/10. Checked it out from school right before Christmas break so I would have a computer at home for a month. In the article here they show Minecraft running at native res and 60 fps. No small feat nowadays considering draw distances that are popular now.

It doesn't run on my 2015 iMac 27" with anything near even medium settings... That's with a Radeon R9 M380, 24GB memory, 3.2 GHz i5. So I find it hard to believe that it will run with max settings / 60 fps on the new MBP, despite the lower native screen resolution.

Believe it.

Re: Apple Silicon M1: Black Magic Fuckery

#619

Post by Hamuko » Wed, Nov 25, 2020, 9:20 AM UTC

piokoch wrote: post #598

michaelmrose wrote: post #511

Earlier quoted context omitted.

A machine that wont outperform the new mini $6000 16 cores 32GB of ram least powerful gpu on the list 2GB of storage $8,799.00 28 cores 48GB of ram same gpu 4GB of storage $14,699 MSRP on a 2020 Toyota Corolla $19,600 AMD Ryzen Threadripper 3970X 32-Core 3.7 GHz Socket sTRX4 $2,629.90 Cost of the same basic GPU about $219 Cost of complete system equivalent to the almost car priced mac about 4200. 9k-15k isn't "a lot"…

Yes and no. 15k hardware is for people who are using it professionally. By "professionally" I mean that they can throw such purchase into their cost and just pay less tax. From my perspective it does not matter if I pay ~15k to tax office or to Apple. In Europe the incentives to buy expensive goods as a company (like cars, fancy office furniture, etc.) is even bigger because of VAT tax (much bigger than US sales tax)…

I'm not sure if a VAT (Value Added Tax) deduction is gonna save anyone's Christmas considering the base Mac Pro is 5320 EUR (6 333 USD) tax-free.

Re: Apple Silicon M1: Black Magic Fuckery

#620

Post by terramex » Wed, Nov 25, 2020, 9:24 AM UTC

aero-glide wrote: post #552

_delirium wrote: post #99

Earlier quoted context omitted.

I wouldn't expect it to be particularly competitive in training large models. It's an integrated GPU with 8 cores, and the "neural engine" has an additional 16 cores. The kinds of discrete GPUs (mostly Nvidia) that people use for deep learning have more like 5000+ cores. I think Apple is aiming more at either training small models, or running pre-trained models. For example Photoshop is starting to integrate neural f…

Is it fair to compare cores like that?

It is not. According to [1] M1 GPU can run "up to 25000 threads".

Comparing raw numbers between vendors is always tricky, but it looks like Apple's "cores" are more like Nvidia's 'Streaming Multiprocessors' (SM's), of which their cards have between 14 and 100. M1 seems to perform similar to their older, mid-end desktop cards (1050 Ti has 6 SM's and M1 matches it in benchmarks).

[1] https://www.apple.com/newsroom/2020/11/apple-unleashes-m1/