Microsoft may be developing its own, in-house ARM CPU designs | Hacker News Reader

Microsoft may be developing its own, in-house ARM CPU designs

arstechnica.com

82 pointselsewhen44 comments

We've just started to clean the mess with browsers, where every browser made a custom custom extensions / API, and now we're entering a situation where every major player will create own CPU with special instructions, and that will cascade to compilers and ultimately to developers trying to provide cross platform apps.

It's sad to see how far the HN community has fallen, to the point in which people don't seem to understand how computers work anymore.

No, this is not a brand new CPU ISA without custom instructions, it's Arm based. And yes, compatibility will be kept with other Arm based CPUs including Apple Silicon.

The M1 has undocumented extensions.

https://twitter.com/dougallj/status/1339934291929694210

I think there's an inherent conflation among people on ARM compatible versus ARM binaries that run on other arm platforms. This isn't the case with x86 today either, so it's weird when people say "oh look extensions mean it won't be ARM compatible". The core ISA is the same, which means the large majority of code will be portable, and people will have to go out of their way to use non portable instructions and intrinsics.

Being an undocumented extension doesn't matter. Compilers already handle per platform extensions on x86.

Being undocumented is better because it means people won't be targeting it in as high numbers, and therefore won't be making non portable code.

The M1 has a completely proprietary GPU. For an intents and purposes the M1 is proprietary.

Did you see the “underwater Stonehenge” article voted to the to the top of the font page the other day? It’s fallen pretty rapidly in recent times. I’m looking for an alternative but have yet to find one. Disappointing.

It often feels to me that it’s the submission titles that get voted up, not the actual content of the linked articles.

> And yes, compatibility will be kept with other Arm based CPUs including Apple Silicon.

Did you forget the three E's?

"Embrace, extend, extinguish"? Surely you're not that young?

I'm old enough to remember going to one of the very first presentations of the Arm architecture in 1985.

As others have said the licensing terms of the Arm ISA prevent this very thing happening and in any event there is no way that even MS could do this given the size of the whole Arm ecosystem.

That's not how licensing Arm technology works.

Now it is how it works for embedded processors[1], I see no reason why that can't "tricle up" to the more powerful ones.

[1]: https://techcrunch.com/2019/10/08/arm-brings-custom-instruct...

Do you think Arm will allow MS to EEE their own ISA?

They allowed Apple to do it, so...

But things will become really interesting at the point ARM would be eaten by Nvidia. I've prepared a big pile of popcorn just in case.

You've answered a question I didn't ask!

Agreed on the popcorn though!!

It’s coming full circle. The period when one company made a processor, a second made an OS, and a third made a compiler looks like it’s ending.

Controlling everything from start to finish like Apple with swift, m1, MacOS can give you a several percent advantage just because of the synergies you can take advantage of.

The PC was the outlier, a serendipity created due Compaq's cleverness, that is all.

> Controlling everything from start to finish like Apple with swift, m1, MacOS can give you a several percent advantage just because of the synergies you can take advantage of.

That's definitely true. Nevertheless, while Apple controls the whole chain for the iPhone, iPhones are not the most popular smartphones in the world. Depending on their use case, rational customers will buy not the best/most optimized, but the one that fits their needs at the lowest price.

Unless one listens to Apple’s marketing, Apple is not necessarily the best and most optimized. In my experience, iPhone’s always feel outdated. They’re often slower, clunkier in their user experience, heavier, and thicker.

Intel has offered a little bit of that confusion with AXV over time. Ice Lake supports all the modern AVX 512 but not the Knights Mill extensions. Previous chips with 512 support have spotty extension coverage. This year’s desktop chips lack them entirely. AMD has had their own issues with SSE extensions they added during the “Heavy Equipment” era being dropped in Zen.

Compiler devs are seemingly awesome at making due with these shifting sands.

Why would anyone want to support programs for Knights Mill on Ice Lake? The (past) theory of Intel that x86 was valuable in itself made no sense anyway. And there is not a single "AVX512" extension set but a base and quite a number of fine grained extensions, and which are supported per uarch is easy to find. One of Intel strategy is runtime dispatch in library code, and for any general purpose AVX512 you are probably going to do that for now. So you could as well address variants if needed. If AVX512 generalizes, I suspect it will become a more stable instruction set. But Knights Mill is legacy already, and has not been replaced.

That’s how the personal computer revolution started. In fact that’s why high level languages were invented. Competition is good. Intel has stagnated the CPU for over a decade now.

In addition we have better and standard instruction sets. Compilers are way more competent as well. If this means faster and more processing power and capabilities we’ll do more than fine.

Won't make much difference in the life of software developers targeting Mac or Windows, unlike browser technology, native app toolkit and languages, api's for these two OS is already unique.

More reasons to use Electron apps. Judging from most of the comments in HN, not something to be happy about.

Previous discussion on the original Bloomberg article: https://news.ycombinator.com/item?id=25472322

"May be?" Isn't the SQ1 their own in-house ARM design?

https://www.microsoft.com/en-us/surface/business/surface-pro...

Microsoft collaborated, but the SQ series is an optimized Qualcomm Snapdragon variant: https://www.itworldcanada.com/article/what-is-the-microsoft-...

SQ1 appears to be designed by Qualcomm exclusively for Microsoft.

I misunderstood then. I thought it was MSFT partnering with Qualcomm to make them a CPU according to a list of design requirements MSFT had defined, including some native CPU-level design work from MSFT themselves.

I hope it's true but I also hope we don't end up with every company selling its own CPU that works only with their own software.

Something like what we have now with x84/amd64 but with more companies on the game.

Well this is how ARM has always been though, you can license an ARM chip and produce your own. That's why you have ARM chips from different companies. The difference with Apple was they are a majority share holder so they probably were able to fully retain rights. Seeing how successful Apple has been with their own chips, I guess Microsoft's taken it as a green light. What's sad is Microsoft's been pushing towards ARM for what feels like forever but it hasn't been up to part with x86. If I can honestly buy a Surface Book with ARM and still get a beefy laptop for a fraction of the cost I will happily buy the most expensive model. I also kind of hope this inspires them to work on Windows Mobile again in the future. They just need it to not lag too far behind Windows 10's core, which I heard was an issue from Windows 7 Mobile to Windows 8 Mobile (there was 0 upgrade path / it was a mess to upgrade).

Apple has never been a majority shareholder of ARM? I think the most they had was 20%, and that is many many years ago.

Pretty close - their initial investment seems to have been 30%.

https://www.latimes.com/archives/la-xpm-1990-11-28-fi-4993-s...

"Apple has invested about $3 million (roughly 1.5 million pounds) for a 30% interest in the company"

Also from that link, this amusing sentence: "Some analysts speculated that Apple might be eyeing ARM chips for use in a future notebook-sized computer"

It was actually 43%

https://appleinsider.com/articles/20/06/09/how-arm-has-alrea...

I can't find any contemporary reporting that says it was 43% - from what I can see, the first mention of 43% seems to come from this - https://www.cultofmac.com/97055/this-is-how-arm-saved-apple-... - which is bonkers wrong about the initial investment ($1.5b!).

Do you have anything pre-dating this nonsense article which says it was 43%?

page 38

http://archive.computerhistory.org/resources/access/text/201...

Edit: Ah I see Cult of Mac quoted Hauser. Well Hauser was a director of Acorn at the time. Also Robin Saxby talks about Apple and Acorn investing $1.5m and VLSI $250k so with some shares for staff that would be about 43%.

tl;dr: 43% might be right.

Saxby talks about £1.5M from Apple and according to https://fxtop.com/en/historical-currency-converter.php?A=150..., £1.5M is just about $3M which I guess clears up the $3M number.

I actually checked companies house but sadly the company (the original Arm Holdings) concerned no longer exists!

Apple was one of the companies that was involved in what is today known as ARM from Wikipedia:

> In the late 1980s, Apple Computer and VLSI Technology started working with Acorn on newer versions of the ARM core. In 1990, Acorn spun off the design team into a new company named Advanced RISC Machines Ltd.,[30][31][32] which became ARM Ltd when its parent company, Arm Holdings plc, floated on the London Stock Exchange and NASDAQ in 1998.[33] The new Apple-ARM work would eventually evolve into the ARM6, first released in early 1992. Apple used the ARM6-based ARM610 as the basis for their Apple Newton PDA.

https://en.m.wikipedia.org/wiki/ARM_architecture

In an ideal world that'd mean we become architecture agnostic and solutions should work on any CPU entering the market while ARM has still not achieved majority support on desktop solutions. Sadly I am certain there will be an extremely painful phase in which some software runs a lot better (more than the instruction set would explain) on some hardware and not on others. Driver support is going to be really poor and tech support will be a very secure job for the coming decade.. hah :)

That definitely looks like the direction we're headed. Hard to see exactly what Intel's business would be in such a world.

Get incorporated into one or more vertical chains and become a “fab” or have a massive architectural breakthrough that maintains the status quo.

This might end up being a big boost for cross platform development.

I’m optimistic about this. If you control everything within your product, then you’re able to fully take advantage of it.

Getting the drivers to work was already a pain, well, this will make it worse it looks like.

Reply on news.ycombinator.com