News [Anand] Arm Announces Armv9 Architecture: SVE2, Security, and the Next Decade

[NTMBK]

Given that Apple has already decided to go their own way with their AMX instructions, and could similarly add other instructions/functionality they wanted, I wonder if they have any reason to go ARMv9?

They released the first Apple Silicon before v9 was finalized, when if they planned on going to v9 as quickly as they went v8 waiting a few months to have that become the baselines for ARM Macs would have made a lot of sense. That seems to indicate Apple doesn't think v9 is that important, so I wouldn't hold my breath waiting for an Apple SoC that implements v9.

If true, Apple's "ARM" could slowly diverge from what the rest of the market is using, at least until external forces forced them to re-base on v9. Microsoft releasing a version of Windows/ARM that requires v9 is the only external force big enough I can think of.

AMX is undocumented. They don't want third parties using it, it is only used in Apple's internal libraries. They could drop AMX from their next chip, replace it with the ARMv9 matrix instructions, and have their internal libraries handle it just fine.

As others have pointed out, there is no reason for Apple to throw away the software ecosystem benefits of staying with "mainstream" ARM.

 

[Doug S]

Lol, you have no idea what you are talking about. When i talk about the ecosystem, i am not talking about apps. I am talkin about frameworks like Node.JS (V8), Rust, .Net, Electron, OpenJDK, Golang, Docker ... the list is endless. As example the only reason M1 Macs have a beta version of OpenJDK is, that there were a Wíndows AArch64 port a year ago and that was heavily based on Linux Aarch64 version of OpenJDK. They wouldn't have Slack, Teams, Discord, VSCode, Spotify and many more without Electron and Node.JS and no chance to get this anytime soon. Docker - not without Golang. And even if you have Docker running - have fun finding Docker images which are anything else than ARM or x86.
Going with an Apple only ISA, they would loose all of the above immediately without any hope to recover from this situation.

The vast majority of that is written in a high level language, the port to ARM is mostly a recompile with very little change required. Back in the day I used to port stuff between RISC Unixes which was WAY more difficult than porting something from one architecture to another on the same OS. The main requirement to change the source code would be from OS level APIs that were supported under x64 macOS/Windows but have been removed under AArch64 macOS/Windows. But that all is pretty quick, the long tail comes from support which requires actual users able to test it and report differences in behavior that result from weird hardware level assumptions that may have been made for stuff like page size or different capabilities of the hardware for stuff like memory protection in stuff for e.g. Java that relies on JIT.

It needs to be worth the investment in time for someone to bother with the port, meaning the port/no port decision is based on customer base. These would have been done due to the fact that the 20 million Macs sold each year were all going to be ARM. I mean, they were willing to port for Windows/ARM which is probably doesn't even have a volume of 2 million a year, so porting for the Mac would have a no brainer they just would have started later. The fact that OpenJDK was already ported for Windows means it is available on Macs sooner, but it would have happened regardless. Same for everything else, if it was supported on x86 macOS it will be supported on ARM macOS. Just not necessarily from day one, but Rosetta 2 does a pretty damn good job until that day arrives.

 

[DrMrLordX]

Thats like Apple shooting themselfs in the foot. They benefit massively from the ARM ecosystem.

If Apple decides to stay on ARMv8.x in perpetuity, they will continue to benefit massively. ARMv8.x is a known entity that is already supported by the frameworks you mentioned in a later post. Given the amount of hardware Apple has brought online with their ARMv8 license, I fully expect Apple to be supported by said frameworks for years to come regardless of whether they adopt ARMv9 at some point in the future.

I don not think it is that niche, considering the amount of Mac users want to run an alternative OS (judging by the resonance in the relevant forums). So they better stay ISA compliant or they immediately lose the ability to run Windows (and Linux).

We're moving into the realm of the anecdotal here. But if you think of Boot Camp users as a percentage of technically-oriented Mac owners who are themselves a percentage of technically-oriented desktop users, you're looking at a percentage of a percentage of people that aren't necessarily a huge part of the overall computing market. Most of those people could be pushed on to VMs should WARM ever switch to ARMv9.

As others have pointed out, there is no reason for Apple to throw away the software ecosystem benefits of staying with "mainstream" ARM.

They don't need to. But there are plenty of reasons not to deal with nVidia.

 

[Thala]

If Apple decides to stay on ARMv8.x in perpetuity, they will continue to benefit massively. ARMv8.x is a known entity that is already supported by the frameworks you mentioned in a later post. Given the amount of hardware Apple has brought online with their ARMv8 license, I fully expect Apple to be supported by said frameworks for years to come regardless of whether they adopt ARMv9 at some point in the future.

No objection. I just had an issues with you using the term "diverge" as in "diverging from ARMv8 without going ARMv9".

We're moving into the realm of the anecdotal here. But if you think of Boot Camp users as a percentage of technically-oriented Mac owners who are themselves a percentage of technically-oriented desktop users, you're looking at a percentage of a percentage of people that aren't necessarily a huge part of the overall computing market. Most of those people could be pushed on to VMs should WARM ever switch to ARMv9.

I am not talking about Boot Camp user, but users, which require an alternative OS including those using VMs - i never used Bootcamp on my Mac but surely i am using VMs. It is also not sufficient to reduce the question to numbers of people leaving the platform. Macs are currently very popular among developers - mainly because you can develop and test application for all major platforms (Windows, MacOS, Linux, Android (via Windows)). Of course you can argue, that developers are just an extremely small fraction of all Mac Users. But if developers leaving your platform, you are going to have issues, independent of how small the population of developers are relative to the population of all users.
And its not just developers, its something called "vocal minority". Its the users very active on the forums, youtube, social networks etc. which have a major impact on how your platform develops by just giving their opinion. Among these people, the ability to run an alternative OS is a very important question.

They don't need to. But there are plenty of reasons not to deal with nVidia.

Not sure what you mean. If both companies have common interests they will deal with each other. Its not like AMD not dealing with Intel because they are competitors - they surely do have common interests.

 

[DrMrLordX]

I am not talking about Boot Camp user, but users, which require an alternative OS including those using VMs - i never used Bootcamp on my Mac but surely i am using VMs. It is also not sufficient to reduce the question to numbers of people leaving the platform. Macs are currently very popular among developers - mainly because you can develop and test application for all major platforms (Windows, MacOS, Linux, Android (via Windows)). Of course you can argue, that developers are just an extremely small fraction of all Mac Users. But if developers leaving your platform, you are going to have issues, independent of how small the population of developers are relative to the population of all users.
And its not just developers, its something called "vocal minority". Its the users very active on the forums, youtube, social networks etc. which have a major impact on how your platform develops by just giving their opinion. Among these people, the ability to run an alternative OS is a very important question.

An ARMv8.6 M2 (or what have you) in the future ought to be able to run WARM in a VM even if WARM begins requiring ARMv9 hardware. There will be virtualization performance penalties, but it will work.

Not sure what you mean. If both companies have common interests they will deal with each other. Its not like AMD not dealing with Intel because they are competitors - they surely do have common interests.

The issue of NV buying ARM has been extensively discussed. Several major ARM players are already fighting the acquisition:

Google, Microsoft, and Qualcomm Don't Want Nvidia to Buy ARM - ExtremeTech

When Nvidia declared its intention to buy ARM, it sent shockwaves through the entire ecosystem. Google, Microsoft, and Qualcomm are particularly concerned, and one of them is said to openly oppose it.

www.extremetech.com

(okay, MS and Google are only in it for the in-house server CPUs but still)

Not sure what Apple's position is, but it's hard to see them as being any more supportive.

 

[Thala]

An ARMv8.6 M2 (or what have you) in the future ought to be able to run WARM in a VM even if WARM begins requiring ARMv9 hardware. There will be virtualization performance penalties, but it will work.

Nothing wrong with ARMv8.6 with respect to run Windows in a VM - at the moment i does not require anything more than ARMv8.2. Again my only concern with your statement was the term "diverge". If the divergence does prohibit downwards compatibility with AArch64, Apple would have a problem, not just with Windows but with the whole ARM ecosystem. Also there is a difference between Windows requiring ARMv9 and Windows supporting ARMv9.

 

[jpiniero]

I am not talking about Boot Camp user, but users, which require an alternative OS including those using VMs - i never used Bootcamp on my Mac but surely i am using VMs. It is also not sufficient to reduce the question to numbers of people leaving the platform. Macs are currently very popular among developers - mainly because you can develop and test application for all major platforms (Windows, MacOS, Linux, Android (via Windows)). Of course you can argue, that developers are just an extremely small fraction of all Mac Users. But if developers leaving your platform, you are going to have issues, independent of how small the population of developers are relative to the population of all users.
And its not just developers, its something called "vocal minority". Its the users very active on the forums, youtube, social networks etc. which have a major impact on how your platform develops by just giving their opinion. Among these people, the ability to run an alternative OS is a very important question.

Windows on ARM has such a low userbase that it's a non factor. If MS goes hard on ARM (and Apple doesn't completely abandon it) I think MS will bend over backwards in supporting Macs given the userbase.

 

[soresu]

Windows on ARM has such a low userbase that it's a non factor. If MS goes hard on ARM (and Apple doesn't completely abandon it) I think MS will bend over backwards in supporting Macs given the userbase.

Typically speaking it isn't MS being supportive that is the problem.

It's usually Apple purposefully attempting to sabotage support for those wanting to run alt OS platforms on their hardware.

The number of big name developers (ala Adobe, Autodesk etc) branching their codebases for ARM Mac should eventually make Windows on ARM more viable regardless of it running on ARM Mac, Snapdragon, Samsung, nVidia or some custom MS SoC solution.

 

[Thala]

Windows on ARM has such a low userbase that it's a non factor. If MS goes hard on ARM (and Apple doesn't completely abandon it) I think MS will bend over backwards in supporting Macs given the userbase.

Not sure why you tend to have issues with grasping an argument. The argument was about MAC users who want to run Windows Applications - which is totally orthogonal to the question of the current Windows ARM user-base, which by the way neither you nor me know.
And then, Microsoft does not support a particular device directly unless Microsoft is the device OEM. Its just that Windows ARM happens to run on many ARM devices and VMs out of the box - provided they have the required UEFI/ACPI.

 

[Thala]

The number of big name developers (ala Adobe, Autodesk etc) branching their codebases for ARM Mac should eventually make Windows on ARM more viable regardless of it running on ARM Mac, Snapdragon, Samsung, nVidia or some custom MS SoC solution.

You don't have to branch at all in order to support ARM. The code typically compiles as is for ARM.

 

[jpiniero]

Not sure why you tend to have issues with grasping an argument. The argument was about MAC users who want to run Windows Applications

But there is so little of a user base of ARM Windows that nobody is going to want to run Windows ARM applications on a Mac since they would have way more software available natively in OSX. x86, possibly, but Apple theoretically forking ARM wouldn't really impact that.

 

[Thala]

But there is so little of a user base of ARM Windows that nobody is going to want to run Windows ARM applications on a Mac since they would have way more software available natively in OSX. x86, possibly, but Apple theoretically forking ARM wouldn't really impact that.

So Mac user do not want to run Windows applications because the user-base is possibly small? Does not make sense at all.
And then of course, Windows ARM does support ARM32, ARM64, x86 and x64 Windows applications (and all ARM64 Linux applications via WSL2) - if the application a Mac user wants to run is available for ARM64 - thats just icing on the cake - otherwise he just runs the x86 or x64 version.

 

[TheGiant]

So Mac user do not want to run Windows applications because the user-base is possibly small? Does not make sense at all.
And then of course, Windows ARM does support ARM32, ARM64, x86 and x64 Windows applications (and all ARM64 Linux applications via WSL2) - if the application a Mac user wants to run is available for ARM64 - thats just icing on the cake - otherwise he just runs the x86 or x64 version.

QFT
IMO the win on ARM is a will be more important then ever
MS shot themself in the foot providing the SW base for apple while having worse performing product (surface)
Microsoft supporting ARM is a big deal for the world

 

[Thala]

MS shot themself in the foot providing the SW base for apple while having worse performing product (surface)

Not sure, why you even comparing a 7W SoC (8CX) with a 15W SoC (M1). Of course the former one is supposed to have less performance. The problem is really people are comparing Apples with Oranges (no pun intended). And this is even ignoring the fact that M1 is 2 years younger and on a more advanced process node.
What the 8CX devices are achieving with only 7W is not much less spectacular than what the M1 is achieving at 15W. You might want to take other 7W SoCs like Lakefield as reference to see what i mean.

ps. This is unless you are referring to another Surface Product like the Surface Pro 7 - here i could somewhat agree to you, as they are also 15W devices but getting destroyed by the M1.

 

[Systems analyst]

Arm vision day.
Segars mentions the ubiquity of future Arm IP.
From [1 trillion machines, by 2035, on] the periphery to the cloud and data centres, and all points in between, including edge computing.

Grisenthwaite mentions:
Armv8 has been the leading Arm standard for 10 years.
Armv9 to be the new leading Arm standard for 10 years.
Armv9 will extend from A-profile, through R-profile, down to M-profile processors.
Machine learning using CPUs, GPUs and NPUs is a growing trend.
SVE developed in conjunction with Fujitsu and others.
SVE2, in Armv9, will extend SVE for use in DSP, ML, 5G and xR [AR,VR] workloads within the CPU.
SVEx will be extended further, during Armv9 development, enabling matrix-based calculations in the CPU, for use in many workloads, SVEx is not just for HPC.
Security is a key issue for computing, to be addressed by Platform Security architecture, (PSA), for use at endpoints.
Arm Confidential Compute Architecture, Realms, will be revealed further, later this year.
Buffer-overflow and use-after-free are persistent software problems.
Memory tagging extensions will address this.
Cheri, a RISC architecture, (from UoC) addresses hardware architecture to improve security. [On MIPS, ArmV8 and RISC-5, with no mention of X86. CHERI is mathematically/logically proven; X86, as far as I know, is not.]
Morello, involving UoC, UoE, Microsoft, Google and others, is a demonstrator board/software project.
If successful, Morello will feed into Armv9 in the middle of this decade.
-----------------------------------------------------------
Arm expect 300 billion Arm powered chips to be delivered by their partners in the next few years. My feeling, as an observer, is that Apple are locked into Arm architecture, and they will implement Armv9.

AWS already use Armv8 in Graviton2, and it seems very likely that they will go on to Armv9, and embrace SVE2, amongst other features. Interestingly, AWS provide AWS Ouposts, allowing from one Graviton2 server upwards, at a client's premises. The main reason given is to reduce latency.
In effect, this looks like HEDT computing in a single, reduced size 1U slot at a client's premises. This must be a direct challenge to Apple's development of M2, M3 etc. processor applications. It also challenges X86.

The introduction of CHERI, a proven architecture, into Armv9 in mid-decade must challenge X86 further. Both Microsoft and Google are involved in Morello, maybe because they see a limited future for X86.

 

[soresu]

The introduction of CHERI, a proven architecture, into Armv9 in mid-decade must challenge X86 further. Both Microsoft and Google are involved in Morello, maybe because they see a limited future for X86.

From what I read of the Morello blurb it does not sound as if it is specific in any way to any ISA despite ARM leading the project in its UK backyard.

Google is involved in Morello for the obvious.

ie Android, Chrome and maybe, soon to be released Fuchsia if the coming developer previews are any indication - all of which are either ARM centric or have ARM ports.

MS is involved in Morello because Windows on ARM will inevitably increase in market share in the future following WARM app releases catalysed by ARM Mac/Mx bringing big name app developers like Adobe and Autodesk to ARM laptop/desktop.

So basically they are involved for the same reason AMD and nVidia weighed in on the development of the AV1 codec, in order to get support for it in their products as quickly as possible.

Likewise MS and Google's direct involvement in Morello should mean a speedy uptake of its benefits.

Though I would not be surprised to see the likes of EV and autonomous vehicle manufacturers like Tesla get in on that action too for security reinforcement of their software stacks - god knows the thought has crossed my mind more than once when SDVs come up.

 

[soresu]

In effect, this looks like HEDT computing in a single, reduced size 1U slot at a client's premises. This must be a direct challenge to Apple's development of M2, M3 etc. processor applications. It also challenges X86.

I'd be surprised if AWS Outposts is at all aimed at the same people.

ARM Mac is clearly shooting for completely replacing x86 Mac in all form factors - Mac Pro to Macbook Air.

That's not what AWS is about, nor is there any indication of Apple directly addressing server type systems again.

At best Apple are targeting a high end, many core ARM workstation for Mac Pro - which is possibly why they have shied away from the clearly superior perf/watt AMD TR offerings vs Intel's Xeon in the current designs so that the ARM Mx solutions seem even better by comparison when they land.

AWS Outposts sounds like an on the premises, ready made cloud/VM server for businesses already familiar with AWS wanting those services with ultra low latency access and a measure of independence from ISP's.

I can imagine that would be quite attractive to businesses not willing to build their own infrastructure, or wary of putting their client data in the wider AWS cloud - which is a very valid concern for medical and other sensitive or valuable information.

After the big gamut of hacking attacks on COVID research facilities I can definitely see a market in this kind of offering.

 

[TheGiant]

Not sure, why you even comparing a 7W SoC (8CX) with a 15W SoC (M1). Of course the former one is supposed to have less performance. The problem is really people are comparing Apples with Oranges (no pun intended). And this is even ignoring the fact that M1 is 2 years younger and on a more advanced process node.
What the 8CX devices are achieving with only 7W is not much less spectacular than what the M1 is achieving at 15W. You might want to take other 7W SoCs like Lakefield as reference to see what i mean.

ps. This is unless you are referring to another Surface Product like the Surface Pro 7 - here i could somewhat agree to you, as they are also 15W devices but getting destroyed by the M1.

I didn't comment devices/hardware, but hardware+whole ecosystem

MS invests a lot into ARM+windows (x86 migration) ecosystem

IMO MS is working in behind on a chip for its cloud servers and laptops, maybe even mobile restart

but Apple gained a lot for free and I am saying that as M1 macbook air owner, if MS didn't invest in the ARM ecosystem that M1 would be a plaything except for the core apple user base

So for free Apple gained a lot and released a very fast laptop- we can be soon in the position as with x86 macbooks, one computer everything covered

that is a big win

about that ARM v9, I wonder what will happen now with the approach of chip designers when Nvidia bought it...

 

[DrMrLordX]

Not sure, why you even comparing a 7W SoC (8CX) with a 15W SoC (M1). Of course the former one is supposed to have less performance. The problem is really people are comparing Apples with Oranges (no pun intended). And this is even ignoring the fact that M1 is 2 years younger and on a more advanced process node.

Correct, 8cx is not a new or current product. It'll be interesting to see exactly how Qualcomm updates it since they have both X1 and A78 cores in the Snapdragon 888. They need to avoid that crappy Samsung node though.

 

[soresu]

about that ARM v9, I wonder what will happen now with the approach of chip designers when Nvidia bought it...

Nothing.

The superiority of ARM Ltd's µArch designs over nVidia's custom ones so far seems fairly clear.

At worst they might steer a particular core in the direction of SDV use, but the Cortex-AE products already target that niche.

 

[soresu]

but Apple gained a lot for free and I am saying that as M1 macbook air owner, if MS didn't invest in the ARM ecosystem that M1 would be a plaything except for the core apple user base

More like Google and Apple invested in ARM with Android and iOS.

There is very little native software on Windows on ARM so far to tempt users sadly.

The ARM Mac moves from Apple have prompted far more movement within the app developer industry than we have seen since WARM was announced.

If anything MS stands to benefit more from ARM Mac happening than the other way around.

The Mac ARM ISA switch forces Mac OS supporting software vendors to either switch from x86 to ARM, or abandon the platform entirely - that is something MS would never do to Windows software developers.

Hence as long as a dev has both a Mac and Windows port we will probably end up seeing a native ARM Windows port not a great deal later than the ARM Mac port.

 

[Systems analyst]

Andrei, if you read these threads, it would be interesting to see an article on CHERI/Morello. In particular, an answer to the question: assuming that CHERI is proven to be correct, and made a part of Armv9, is this a key distinction between CISC and RISC? If it is, will this lead to the demise of CISC architectures in general, and X86 specifically?

 

[Thala]

I didn't comment devices/hardware, but hardware+whole ecosystem

MS invests a lot into ARM+windows (x86 migration) ecosystem

IMO MS is working in behind on a chip for its cloud servers and laptops, maybe even mobile restart

but Apple gained a lot for free and I am saying that as M1 macbook air owner, if MS didn't invest in the ARM ecosystem that M1 would be a plaything except for the core apple user base

So for free Apple gained a lot and released a very fast laptop- we can be soon in the position as with x86 macbooks, one computer everything covered

that is a big win

about that ARM v9, I wonder what will happen now with the approach of chip designers when Nvidia bought it...

I am not denying that Apple got a lot for free. But Apple is gaining this, because they are part of the ARM ecosystem - this means the effort one company is putting into the ecosystem might be benefitting others. But thats the wonderful nature of an ecosystem. Its not like MS is not benefitting either. I think the release of Adobe Photoshop and Lightroom for Windows ARM64 was tightly related to the release of the M1 version. And then Apple is benefitting from Microsoft regarding OpenJDK - as Microsoft put much effort last year into OpenJDK Windows ARM64 port - which the M1 version is based off.

Coming back to devices. When MS started with Windows ARM, there were just servers in the 200+W power range and phone chips in the 3-4W power range. I am totally happy, that Qualcomm with Microsoft were at least shooting at 7W, when the 8CX was announced.

 

[Thala]

Correct, 8cx is not a new or current product. It'll be interesting to see exactly how Qualcomm updates it since they have both X1 and A78 cores in the Snapdragon 888. They need to avoid that crappy Samsung node though.

Yup and if you read the phone reviews, you come to the conclusion, that 3W for the whole phone (that is where the Snapdragon 888 power consumption stabilize under load - and Andrei is measuring device power and not only SoC power) is iust not enough for the X1 and a powerful GPU under load combined with the Samsung node.

 

[Doug S]

So according to the release notes for ARMv9 SVE2 is optional in ARMv9. According to this ARM document about SVE both SVE and SVE2 are optional in ARMv8.

In fact, it looks like all the major new features in ARMv9 are optional. I can't find anything listing what has definitively changed in ARMv9 to stop you from calling an ARMv8 CPU ARMv9. I'm sure there are some things (maybe side channel attack fixes?) but nothing major as far as I can tell.