Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Linux running natively on M1?
- Thread starter robd003
- Start date
-
- Tags
- linux
- Sort by reaction score
Internaut
macrumors 65816
This isn’t too surprising. The M line potentially provides a route into servers (or at least a critical bit of the IP behind them) for Apple. But macOS isn’t, and probably won’t be, an enterprise or carrier OS. Only Linux fits the bill at the moment. I would be surprised if the likes of RedHat, Amazon and Ubuntu didn’t have some sort of line of communication with Apple.
That comes at an interesting price, though. The way that Apple is working with Asahi on this, would by extension mean that some of the architecture for Silicon to be supported by Linux has to make its way into the kernel. By the time that happens, that has to conform to the preferred licenses used for the kernel (GPLv2, LGPLv2, etc.) At that point there is nothing to stop other companies from looking at that code and using it (should they reference the original authors/developers of the code), or re-engineer the code to make it proprietary to their company.
This could mean that if Silicon support hits the kernel, NVidia or Intel could take the code and build their own CPU based on that architecture, and create a competitive processor against Apple. Microsoft, Sun/Oracle, IBM, and HP could do the same, and as they already have data center/enterprise offerings, Linux may be their key to getting their hands on Silicon-based code.
BL.
Even if we assume that Apple is interested on the server business (which is far from obvious), M1 can already run Linux - via virtualization. It’s a straightforward, cheap, and robust solution. There is no reason whatsoever for them to mess with direct Linux booting or drivers or kernel patches. MacOS even includes a ready to use framework to virtualize a Linux kernel.
Apple is not working with Asahi on anything.
Asahi folks already submitted a plethora of patches to the Linux kernel. The code is already there.
What would be the point of such an enterprise? How would any other company benefit from adopting Apples non-standard interfacing protocols?
How would some interfacing code allow Nvidia or Intel to create a competing CPU? It provides zero insight into the basic CPU architecture or how Apple is able to achieve their superior perf/watt.
You may want to tell that to that tweet.
Which proves my point.
In my time working at Intel, I saw how Intel purchased CPUs from AMD to see how they worked compared to theirs, and to the point of reverse engineering some of what AMD did to make a better CPU, and even adopted some of what AMD did. The results of that were the Core family of CPUs. They still even do that now with Ryzen.
You may want to ask AMD that. I mean, as they wrote the x86_64 spec and interfacing code that Intel adopted, Intel created a competing CPU that for a fair while beat the pants off of anything AMD could offer. Why do you think the Core series has been popular for Intel over the past 14 years? Rana wasn't able to beat Sandy Bridge or Ivy Bridge. Nothing AMD had could have competed with Haswell or Broadwell at the time. AMD finally caught up after Sky Lake, Comet Lake, Cascade Lake, and Ice Lake started to have major heat problems. AT that same time was when Apple decided to go it alone and dump Intel.
If Intel could do that with AMD (hell, Cyrix did that with Intel!), then Intel could do that with Apple.
BL.
And yet, Apple is not working with Asahi on anything. It was never the case that some Apple engineer contacted Asahi people and offered them documentation or did code reviews for them or asked what kind of feature they want to see. These new features are great for Asahi folks, sure, but it could be just Apple streamlining their basic interfaces, or maybe indeed some higher executive on Apple said, hey, let's throw a bone to the FOSS community. But implementing a feature that makes some stuff marginally easier is far from "working with someone".
Studying your competitor's products and learning from them is standard practice. What does it have to do with Linux kernel code? They are not studying the interfacing details (those are meaningless for the purpose of building CPUs), they are studying the CPU's behavior (instruction scheduling, latencies, reorder buffer sizes, branch predictor performance...). All this information have been available for M1 since early summer thanks to the efforts of researchers. And I am sure that both Intel and AMD have their own testing labs where they dissect M1 as much as they can.
Intel Core is an evolution of Intel P6 (Pentium Pro). What does it have to do with Ryzen?
How exactly do you think CPU software interfacing details like interrupt controller interface or proprietary NVMe variants could help Intel to reverse engineer Apple CPUs? That's like reverse-engineering the car engine from a picture of a steering wheel...
Definitely not because Intel copied some tricks from AMD... AMD had great success with K7 since Intel got stuck with Netburst, an microarchitecture that in hindsight did not fulfill the expectations. The result of this was that Intel went back to the drawing board and took an older architecture (P6), which prove to be much more scalable than Netbrust (or K7, for what matters). I mean, current Alder Lake Golden Cove is still based on the 1995 Pentium Pro.
Intel didn't "do" anything with AMD. Intel simply had a superior microarchitecture and better execution. Now Apple has a superior microarchitecture and better execution.
I'm really interested if Apple is gonna make servers for themselves using ARM or if they going to hit private the private clouds and then eventually public ones.
I with with Azure and we are looking into migrating every non-x86 based workload to ARM as it's cheaper and we get more bang for the buck.
I with with Azure and we are looking into migrating every non-x86 based workload to ARM as it's cheaper and we get more bang for the buck.
Moore law is dead, a popular leaker of PC hardware, has confirmed that Apple is designing its server chips. It seems that none else has corroborated it, so it's up to you if you believe him.
Are Azure's ARM-based virtual machines better than AWS' ARM-based virtual instances?
What do you mean by "better execution"? Use TSMC 5nm node? I thought a microarchitecture is a physical implementation that uses a particular ISA.
Why would Apple need help from other companies to make the Apple SOC use Linux-based OS? Apple could create its Linux-based OS as Microsoft has done. Besides that, the trickiest part is the drivers, and none knows better the Apple SOC than Apple itself.
What makes macOS unsuitable for cloud servers?
Mostly the fact that Apple has been able to maintain steady flow of micro architectural updates.
What makes macOS unsuitable for cloud servers?
It’s not that it’s unsuitable per se, but the standard cloud server infrastructure and tooling is too Linux oriented. I am sure that a stripped down Darwin would perform reasonably good as a server platform, but setting it up and maintain could be a problem. I think virtualized Linux containers could be the simplest way to go if Apple wants to get into that business, but I’m not sure that they would - it’s risky and very very competitive. Apple likes products with high margins.
P.S. Also, Phoronix tests are quite useless - most of the time. Tests where Linux does significantly better can be usually attributed to OpenCL drivers, obscure software that doesn’t play well with Mac, or the fact that Linux by default does not do fikesystem event tracking (that’s for example where the big wins on git etc. come from).
Last edited:
Internaut
macrumors 65816
Why would Apple need help from other companies to make the Apple SOC use Linux-based OS? Apple could create its Linux-based OS as Microsoft has done. Besides that, the trickiest part is the drivers, and none knows better the Apple SOC than Apple itself.
What makes macOS unsuitable for cloud servers?
It's a compelling argument you make but I’d caution there are quite a lot of competencies involved in an enterprise or carrier grade OS. Mainly around local and geographic resilience, bonded (i.e. very hot standby) network connections, disaster recovery and so on.
Do you suggest that Apple should use a type 2 hypervisor instead of type 1?
I think cloud services could harm premium laptop sales. Why would I buy a maxed spec MacBook Pro if I can do the same with a MacBook Air connected to the cloud?
What makes you think that Apple doesn't have that knowledge in-house? After many years of using Linux for internal servers, it should be easier for Apple than for Microsoft to create its in-house Linux distribution because macOS is Posix compliant.
Internaut
macrumors 65816
What makes you think Apple do have the experience in question? They’ve never shown any interest in the Enterprise/Carrier space. Whatever they do in house is likely very specific to Apple’s needs. The likes of Amazon, Google, Microsoft, Oracle and RedHat (not to mention NSN, Ericsson and Huawei in the carrier space) have vastly more experience delivering to a diverse range of clients.
Not suggesting anything. Just pointing out the path of the least resistance.
Not sure I follow. What kind of services do you think that Apple will offer that could replace what you do on your laptop? I though we were talking "cloud services" as in AWS or similar?
Why would Apple bother with creating an in-house linux distribution?
After reading marcan, I can understand you now.
Other cloud service providers offer computing power. For instance, I can connect to a more powerful computer in the cloud when I need extra computing power for data science/ machine learning tasks.You can connect to a virtual machine for data science/machine learning tasks or a render farming for render tasks.
Yep, exactly.
Ah, but these things already exist and they do not obsolete the need for powerful laptops. We really run pretty much all data analysis on a supercomputer anyway, but you first have to develop/test locally. These things go hand in hand. Besides, remote offloading of computation is not a panacea. It's great when your workload needs hours or days to run, but what if it's just a couple of minutes? Long enough to be annoying over time, not long enough to deal with the overhead of remote work submission.
Many organizations use Excel as a required part of some of their internal applications or workflows. I haven't worked at a company in the last 30 years that didn't use Excel for something important. The MS Office suite is the standard in many large corporations.
I think it is sensible not to tinker with a system you are using to get work done so I have zero interest in tinkering with my Mac unless its on a VM.
This was in response to a particular use case, where it makes very little sense to use excel. This was not a comment on excel’s utility or pervasiveness in industry.
You probably never had to fill out an EU research council time tracking form
One of my colleagues ended up writing a script that would just fill random times adding up to an expected weekly average. Those timesheets were one of the most ridiculous things I have ever seen.Speaking of porting Linux to unusual hardware, here is an interesting talk about porting Linux to a PS4. Not Apple hardware but some similar problems. What is interesting is how similar to a T2 Mac the PS4 architecture is. It runs a BSD Unix based OS on x64 with an AMD GPU and has an ARM coprocessor also running a BSD based OS (like the T2).
If Linux is really needed why not do like every normal person and just run it in a VM? MacOS even comes with a system framework for running Linux.
This was always the case for me. Even on previous Windows Machines, it worked, but driver support was terrible.
But that would be the easy solution, and what's the fun in that? LOL