Hacker News Reader

Comment by naderkhalil | Hacker News Reader

> Will Microsoft now offer an official Windows 10 on ARM that can run on M1 Macs?...This could create trouble in the x86 PC world. Or will Microsoft, less committed to Windows than it is to its applications running everywhere, be content to see Office and more run (and generate revenue) on M1 Macs…and how will HP, Asus, Dell and others react? This is just the beginning of a new competitive layout.

Could this be good news for Linux?

View on HN

Why? If everyone gives up and just builds everything on the web then it's an indirect boon for Linux, but M1 will probably never ever properly support Linux.

Unless Apple have a complete change of heart, if it ever does it will be as a result of far too much work done for free

Early days, but Hypervisor.framework patches are being built that allow QEMU to run an Arm build of Linux (and Windows) at native speed.

https://twitter.com/jonmasters/status/1333555759737663489

The reality is that some of the drivers for the M1 chip will never arrive, but that if you want to run Linux (or Windows) Arm code then you can do so soon.

The issue for me is that I'd love to have an ARM laptop but I also love playing with the hardware and measuring it, trusting the hypervisor makes that a lot harder.

If I was a web developer I'd probably buy one (not a knock on web developers, if I was any good at it I'd do it), but I like playing in the grey area of the Venn diagram where microarchitecture, compilers, and kernels meet.

Wouldn't these be QEMU patches to let it support Apple silicon Hypervisor.framework?

I was a little confused by this part because MS is already happy for Office to run everywhere, in browsers. Given the in house expertise that’s given us VS Code I wouldn’t be at all surprised to see an Electron powered Office app suite in the not too distant future. Linux would (and does) benefit just as much as Mac.

I think it will benefit Linux but not that way. It will force the x86 vendors to adopt the microarchitectural aspects of the M1 that make it fast in practice, instead of spending so much area/power budget chasing SPEC benchmarks. We will get better chips through competition.

I don’t think these goals are exclusive. I am not sure about the M1, but the A13 and A14 are doing quite well in SPEC benchmarks. If they have compromised these scores, it’s not obvious to me.

I don't think they've compromised on SPEC performance at Apple, but I think they've chased it too hard at Intel. I'm sure I'd rather have more iTLB entries and more L1i cache than that third 512x512 FMA unit, if that's a trade that can be made.

Sorry I misinterpreted. Isn’t this a reflection of the primary markets for both companies, though? When Intel is trying to sell chips for HPC clusters and expensive workstations, impractical vector units can make sense. Clearly Apple is not bothered by that and has a more balanced micro architecture for more common use cases.

Second, not third, and chasing benchmark scores (instead of real workloads) would result in exactly the trade off you’re asking for; most pieces of SPEC will barely benefit from the second AVX512 FMA at all.

It will force the x86 vendors to adopt the microarchitectural aspects of the M1 that make it fast in practice, instead of spending so much area/power budget chasing SPEC benchmarks. We will get better chips through competition.

There's only so much you can do with a 40-year old instruction set. There's certain things neither ARM or Intel can do compared to the M1.

Example: the max number of instruction decoders Intel and AMD can use is 4 and that brings tons of complexity due to x86 instructions ranging in size from 1-15 bytes. There's no way for the processor to know where the instruction boundaries lie except through trial and error essentially.

All ARM instructions are one size, so decoding them and having many instructions in flight at any given time is much easier. The M1 has 8 instruction decoders.

Short answer: the M1 can process twice as many instructions per clock cycle than Intel/AMD processors can.

The M1 and the Zen3 run neck and neck, with the Zen3 being slightly faster in some benchmarks.

But the Zen3 runs at 5 GHz and the M1 runs at 3.2 GHz, meaning if the M1 ran at 5 GHz, it would blow the Zen3 away. And of course, on a performance per watt basis, the M1 is the obvious winner.

This article gets into all the reasons why Intel and AMD are between a rock and a hard place when it comes to competing with the M1:

"Why Is Apple’s M1 Chip So Fast?"—https://debugger.medium.com/why-is-apples-m1-chip-so-fast-32...

It sounds like you are just regurgitating some blog post you found. I don't have any workloads that are decode starved on x86 so it wouldn't make any difference if Intel had 50 decoders. I do have tons of workloads that spend a lot of time on TLB misses. I have several that are cache fill limited.

It’s not about being starved for anything; it’s that Intel and AMD have legacy issues they can’t get around. They’ve pretty much out of bullets: more cores, higher clock frequencies, better manufacturing, etc.

The point they can’t process as many instructions per clock cycle and there’s nothing they can do about it. They can’t match the M1’s performance per watt.

And certainly there’s no Intel/AMD laptop processor with 8 or 16 Gb of RAM via a 128-bit memory bus on the same die. Same thing with a 16-core Neural Engine with an OS with the APIs to take advantage of them.

Intel's current generation CPUs have the same dual-channel LPDDR4X memory that Apple uses. Memory is not "on the same die" for either company because you don't make DRAM on the same process where you make CPUs.

The grandparent used the wrong term. The memory isn't on the die, but it is in the package (MCM). I don't think that provides faster throughput, but it should help latency.

Regardless, the big win was giving unified access to that memory using a memory fabric. That lets the GPU (for instance) access graphics assets with no copying - it just uses them in place.

AMD can absolutely do the same thing, if it desires. Intel can too, but it has a hill to climb WRT GPU performance.

Apple's AI chiplet is another matter...

Part of this reflection is weird.

Office is already running on Mac OS. M1 does not change anything in that regard.

So the only meaning that remains that would make sense is: how would HP, Asus, Dell, and others react if MS sold non-OEM licenses for Windows 10 for ARM.

Given for now I suspect they don't ship a lot of ARM laptops, that boils down to; would the M1 hardware be unfair competition.

I don't think so. MS x86 emulation is likely inferior to what Apple is doing for now, and I'm not sure if/how quickly they could make use of the M1 specific capabilities if they even wanted to. So at least the market won't shift quickly.

Will it shift slowly? Maybe.

I was thinking about the same thing.

If the commitment of Microsoft is in question, there’s a good chance of increased resources by Dell, Asus etc towards Linux.

Reply on news.ycombinator.com