Hacker News Reader

Comment by mindfulhack | Hacker News Reader

Apple Silicon feels like the most exciting 'Apple computer' news since the launch of the Macintosh.

Typing this from a 16-inch MacBook Pro that cost me $5k, I know Apple loves to make money via raw sale price of premium hardware. Imagine what power they'll be able to pack into such a price once they apply Apple Silicon to the MBP. I'm thinking 128 or even 256 GB RAM, extraordinary processing speed options that are orders or magnitude faster than any other laptop (or even desktop), it's major major news.

The big question mark hanging over it is ability to run native Linux. I'm a Linux user, increasingly. We'll have to watch this space.

View on HN

> extraordinary processing speed options that are orders or magnitude faster than any other laptop

If ARM-based chips can be "orders of magnitude" faster than x86, they'll already be dominating datacenters. The performance improvement in sustained compute (within a power envelope) will be incremental rather than the giant leap you're hoping for.

x86 CPUs do not execute their CISC instructions natively; they get decoded to simpler instructions. But the decoder consumes only around 5% of the power budget, so it's not like x86 is spending significant amounts of energy because of the legacy instruction set. Now, due to a decade of high-stakes innovation in the mobile market, ARM solutions will do significantly better in idle power consumption (which was essential for phones).

So, I'll expect way better standby and well managed low-power states/transitions. But a minor increase in full throttle power.

I haven’t been at Intel in ~8 years, but your description of the implementation of x86 is nothing like the processors I worked on. Admittedly, I never worked on the Xeons; but I would be surprised if decode were 12 watts on those platforms. My guess is that’s off by an easy 2 orders of magnitude.

My information is second-hand (I am a software guy); so feel free to correct me here for the benefit of other readers.

But here's a study which arrives at a number between 3-10%: https://www.usenix.org/system/files/conference/cooldc16/cool...

Yep, looks about right. That benchmark is designed to maximize the decoder cost of the CPU, while effectively idling the rest of the CPU. On some of the architectures I worked on, with the thermals turned off, you could decap the part with decode. But you could also do that by jamming the x87 unit, issuing L0 prefetch, etc. The same is true on the ARM parts I work on. In practice, though, user code doesn’t do this. I’d be interested to see this when the larger L$ are lit—they stop at L2.

Also, mops aren’t “RISC”; they’re mops. The only part of the x86 parts I worked in that used a decode ISA was the network & DMA, which desugared down to x86i.

> If ARM-based chips can be "orders of magnitude" faster than x86, they'll already be dominating datacenters.

Raw compute performance is the wrong metric for measuring data center worthiness. The correct metric is performance per dollar.

ARM in the DC currently suffers a lack of ARM-specific code optimizers and has to play catch-up in this regard. Amazon recently focused a squad of engineers on adding ARM 64 optimization to the Zend Optimizer in PHP 7.4 to an incremental effect, but then as some cruft is removed on the PHP 8 roadmap, that gain is around 15% in raw performance. Combine that with 20% lower pricing for comparable Graviton2 EC2 instances, and you’ve got a significant cost motivation to migrate to the ARM platform.

https://aws.amazon.com/blogs/compute/improving-performance-o...

Expect AWS to dedicate engineering resource to similarly bring projects like Python, V8, LLVM, etc. into the ARM-future.

Note that in performance per dollar, the dollars include cost of the parts, fractional cost of rack space, watts consumed directly and watts consumed indirectly for cooling.

There’s also the fixed engineering costs for the life cycle of new hardware and for customers performing migration (where relevant).

That's interesting - do you have a source for 5% of the power budget spent on decoding?

Sorry, I can't recall where I read the 5% number, but here's a study which investigates the impact: https://www.usenix.org/system/files/conference/cooldc16/cool...

They put it at somewhere between 3-10%.

Thanks - and of course ARMv8 decoding will have some cost. I seem to recall that x64 has greater code density too so some offset there.

Directly booting into Linux sounds unlikely by what Federighi said. MacOS would be around as a hypervisor at least, but he claimed that this technology is efficient enough, that it doesn't make a practical difference.

I am doing my day-to-day work with Linux in a VMware VM on my MB Pro and couldn't be happier. When running fullscreen, I don't notice any difference to having a "native" Linux install, plus I am getting all the benefits of being able to suspend my VM, having several VMs and of course, switching to the macOS desktop is just a two finger swipe on my mouse. This really has become my favorite compute setup.

The big benefit of using VMware for this is, that the necessary "drivers" are open source and have been part of the Linux kernel for quite some time. So you don't need to install custom drivers in your VM any more. Expecially with very new Linux kernels this means, you don't have to wait for the VM provider to supply new drivers. Which is the one thing I didn't like about Parallels, it rarely supports the latest Linux kernels.

I'm not a dev, but for my particular use case this is my current observation:

Running Windows or Linux as a VM in macOS has terrible performance in VirtualBox right now (people are discussing it at length in VirtualBox forums). The expensive software Parallels is the only way for it to be usable. VMware is maybe OK, I don't use it much and only briefly tried it the other day.

It seems if you want to run Windows or Linux as a guest VM, it's better to make Linux your host OS than Mac. I also increasingly prefer open-source software, so VirtualBox at the very least is desirable. On macOS, that solution is not adequate right now.

Virtual box has always been a sub par experience on both Mac and Windows. Parallels is the best.

I stopped doing VMs and just ssh or Remote Desktop to my Linux and Windows boxes. For a little while I was running a small ESXI server.

Virtualbox is shit, agreed. Try Fusion, specifically the Technical Previews. They are free, and run like a dream(you can turn off debugging). https://blogs.vmware.com/teamfusion/2020/07/fusion-big-sur-t...

> Running Windows or Linux as a VM in macOS has terrible performance in VirtualBox right now (people are discussing it at length in VirtualBox forums).

That's my experience as well. Running macOS VM is even worse. My understanding is that it's because all of these rely heavily on 3D acceleration for the GUI and VirtualBox's acceleration support has not worked properly for a few years now. It's slightly better if you run VirtualBox in the low resolution mode but still not ideal (and of course the graphics are... low resolution). It's quite obvious that macOS hosts are not a priority for Oracle, I mean, a few VirtualBox releases couldn't even be installed (missing notarizations and one other occasion I cannot remember right now) and Oracle noticed only after people started complaining on the forums...

These says I use VirtualBox but in the headless mode and I connect to it via SSH and VS Code's SSH remote extension.

I only had good experiences with VMware on the Mac. You should give it a try. I prefer it to Parallels for the listed reasons.

I’ve been using VMWare Fusion since version 3 or so for work running Linux dev environments. It has generally worked great, which is more than I can say for VirtualBox. I do hope there will be a way to run x86 Linux VMs on Apple Silicon but not holding my breath.

Fusion Pro doesn’t have to be expensive either. It’s easy to find $5 gray market keys on eBay.

Wouldn’t know about that since I have no reason to pirate professional software, my company pays for the tools I need to do my job.

This is how I feel. It would be the first Apple hardware for me in almost two decades. It's definitely the nerd factor of the idea of having a well designed non-x86 laptop with good specs, but we're currently planning to evaluate arm based servers, so having that as your dev machine doesn't sound like the worst idea.

But I'm also exclusively using Linux nowadays, so I guess we'll have to watch this situation unfold for a while to see how viable running Linux on those is. I'm pretty sure it will be possible, but I'm not sure how much pain it will be.

I'm pretty encouraged at how far the Linux kernel's progressed with adding support for current modern MacBook hardware.

Every now and then I boot up a distro that tends to have quite bleeding edge kernel version like a Tails USB to see how complete the hardware compatibility is out of the box.

Apple's hardware like their trackpad remains superior enough that there seems to be enough interest (like you and me) in seeing it work.

https://github.com/Dunedan/mbp-2016-linux

I don't know if anyone else has caught up on this in the last two years since I went laptop shopping, but the trackpad quality is one of the only major things keeping me from buying a Windows laptop and installing Linux.

People bring this up, but honestly I’ve never noticed trackpad quality issues on other higher end PCs.

People are basing their OS choices on marginal difference in trackpad quality?

How about the fact that with macOS your tied to one vendor and their hardware choices (keyboard /ports etc). I don’t have a new MacBook(work machine 5yrs old), but I didn’t love the giant trackpad on the one I borrowed.

As an example of a hardware issue on the XPS 13, the trackpad is hinged from the top near the spacebar.

That means that it either requires more force to click the higher up that pad you go or you have to click more precisely at the bottom of that pad.

That's been a non-issue even before they moved to the artificial click mechanism.

Me too. I've never been a 'fanboy' of any given manufacturer or ecosystem. Even while still a Windows user I starting moving my laptop hardware to Apple in 2010 once I saw it clearly was the best. I painfully lived through the awful #keyboardgate years in 2017-19 and lived to tell the tale, and I'll only move to a different 'whole package' that's superior as soon once someone actually makes one. I recently realised maybe trackpad isn't everything, and I guess it's not. It's about the balance of the whole package. Since I adore macOS anyway (it's like a half Linux what with homebrew), I have no problem with it for now.

As of a couple of years ago, last time I checked, macbook’s wifi stack was totally nonfunctional on Linux. Has that changed?

Update: I see from your link it works on 2/8 tested. Better than nothing.

How would make ARM chip DDR RAM cheaper? I expect Apple to finally fix cooling and power management, but that is all.

By saving money on CPU, they can now pack in more RAM to keep the price up, and give the machines a competitive edge.

RAM Prices on Mac where never reasonable.

Upgrade prices where never reasonable.

Why would Apple ever reduce upgrade prices just because they might save 100$ on a CPU?

And why would you assume that the development of your own inhouse ARM CPU would make such a financial difference? After all, you have to develop a CPU.

And what? Apple reports over 25% profit; Why would you even assume that Apple is doing anything for you? They could easily reduce there profit margin for you. They don't.

It's a fallacy to believe that costs have anything to do with prices, except as a lower bound.

I wonder how much they’ll save. They sell millions of cpus and can distribute the engineering cost across them, but compared to amds/intels processor numbers it seems small.

Plus it seems that the fact that they now need developer to write video drivers/ compiler optimizations etc are all additional costs.

They payoff can be big, but so are the risks.

Anyway Apple charges too much for DRAM capacity CTO option so Apple could cut prices now if they wish.

I think we will see custom acceleration hardware in the future to support stuff like video editing

Don't most computers already have hardware accelerated encode/decode via cpu/gpu?

The real point is, Apple is an integration company that has now added self-made silicon to its toolset. Whereas in the past, they created the M chips for tracking motion in phones and watches, and the T chips for Secure Enclave and Secure Storage, the roadmap Apple displayed during the keynote indicates they are going to create specialty chips for any application that makes sense.

How big is that market?

Big enough that they already sell an fpga card for the Mac pro to do it.

Huge. Everyone edits video now.

Yea, not in a way that requires specialized hardware though. Edits on the phone are already virtually instantaneous.

… because of the specialised hardware and the tiny video resolutions supported.

Tiny?

2688x1242 isn’t tiny.

Neither is 4K at 60fps.

Without color grading or other usual video correction.

Once you start to tinker with your images, the picture change quite a lot for the cpu/gpu

>'Imagine what power they'll be able to pack into such a price once they apply Apple Silicon to the MBP. I'm thinking 128 or even 256 GB RAM ...'

I'm interested in why you think Apple will be able to build an integrated memory controller for a laptop that is capable or 128 Gigs or 256 Gig of RAM. Most laptops with Intel were stuck at 16 Gigs of RAM for years although more recently that has increased. My understanding about the reason for this is that these high capacity integrated memory controllers are very difficult to design. Is there something specific to ARM that would make iterating on these easier for Apple than it has been for Intel?

Apple learned (prior to 16”) that a large number of customers want to run huge VMs, load massive AI models and have faster imaging performance that comes with higher memory limits. I don’t think they are going to ever want to introduce another Prosumer machine that people slam publicly that hard again, plus they can’t exactly use the iPad Pro memory controller so they are designing a new one anyway.

Isn't it possible to buy most high-end smartphones with 256 gig of RAM?

That’s storage, not RAM. 8GB of RAM is still a large amount for a phone.

> The big question mark hanging over it is ability to run native Linux.

I wouldn't not be surprised if Apple took every measure possible to make it impossible to run Linux on their new machines.

It is not clear to me, why they would care much about it. The selling point of Macs is to a large amount macOS. On the other side, at least for now, it is announced that Apple Silicon hardware only boots into signed operation systems, which currently is only macOS. I would assume, that means there is little chance to run Linux natively on Apple Silicon. On the other side, virtualisation technology is part of macOS, so there should be good support to run Linux inside a virtualized environment. They even showcased Linux during the keynote.

The only selling point for many developers is that you need a Mac to compile a program for iOS.

Apple would have never gotten any of my money if it weren’t for that.

I am a developer owning a Mac and having never written an iOS application in my life. So yes, that you have to have a Mac to do iOS development did add customers to the Mac platform, but there are many more reasons to own a Mac. For the general consumer it is an attractive platform too, as it is still very easy to use and maintenance is minimal.

I don’t get the “So yes” part. If you develop for non-Apple platforms on a Mac, it is because of X, Y and Z not because you have “never written an iOS application.” As far as a “general consumer,” they had all be using iPad, Dex or Chromebook by now.

The "so yes" refers to the fact, that some of the Macs are bought just for iOS development, but not the majority. I see a lot of people using Macs just as a general computer and a lot of those Macs are laptops of course. I have not even seen a Chromebook used yet and also don't know anyone using an iPad for typical computer work.

we'll see.

for the moment, most people i know complain about poor performances of linux vms in mac os, particularly when running docker (that gets pretty i/o intensive when dealing with images).

> at least for now, it is announced that Apple Silicon hardware only boots into signed operation systems

this alone already rules out gnu/linux i guess?

It really depends. After all, there are UEFI x86 systems using secure boot to boot Linux, but there was also Microsoft's vision of ARM (spec outright prohibits user managed keys, and nobody's shipping with anything but Microsoft keys).

> but there was also Microsoft's vision of ARM (spec outright prohibits user managed keys, and nobody's shipping with anything but Microsoft keys).

Indeed. "Microsoft <3 Linux" they say. Right...

Yes, unless Apple signs some boot loader for Linux, that would rule out Linux booting.

Secure Boot can be disabled on Arm macs...

Do you think that or do you know that? The hardware is unreleased yet, so how do you know?

Apple explicitly said it during WWDC sessions. (this one: https://developer.apple.com/videos/play/wwdc2020/10686)

They don't need to spend any intentional effort on this. There's already the massive obstacle for a good desktop experience: no open GPU driver for Apple GPUs.

Even before that, there's the question of how they enumerate devices on the platform, how standard or custom the peripherals are…

Yes, they are going to spent a lot of money to prevent their customers from installing Linux. All two of them.

FWIW they haven't even invested the tiniest amount of effort into stopping people from installing MacOS on "Hackintoshs". And that practice is probably more widespread than Linux-on-Macbook, and has a more obvious theory of revenue impact.

(It wouldn't be possible to completely make it impossible without "trusted computing" hardware, or, coincidentally, switching hardware platforms to something not available on the generic market. But they certainly could think of a new trick to make life miserable for these users with any point update, and yet I found MacOS offered better support for my random hardware than Linux, without even trying.)

The end effect may be (almost) the same, but I see them make no effort to make it possible to run Linux on their new machines, not take “every measure possible to make it impossible to run Linux on their new machines”.

Thevy've already said there will be no Bootcamp, my guess is running Linux on Apple Silicon is as likely as running Linux on an iPad.

Did they not already commit to running VMs on Big Sur on their ARM chips? ARM-based Linux exists and will run full speed. Because they are kicking kext’s to the curb and stepping up with their own virtualized hardware drivers, we may not have to keep waiting decades for VMware or others to pull their head out with respect to 3D support. The best GUI apps already run on iPad and all of them run on ARM Macs so I doubt new Apple is thinking like they need to hobble the VMs to prevent Linux GUIs from taking off.

But booting these Macs into anything but BigSur is not happening as accessing the hardware without using Apple’s drivers could cause product returns.

Reply on news.ycombinator.com