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

[Hitman928]

ARM has announced ARM v9. Anandtech article below.

Arm Announces Armv9 Architecture: SVE2, Security, and the Next Decade (anandtech.com)

 

[moinmoin]

Hm, aside making SVE2 a part of ARM proper (yay!) the focus on security interest me the most, so I'm very disappointed how non-informative the announcement appears to have been in that regard. My hope is that it catches up to AMD's flexible approach to per instance/process memory encryption as well as I/O isolation, though the barebone slides shown don't really indicate any of that.

 

[soresu]

Welp, called it.

 

[DrMrLordX]

So SVE2 really isn't a part of Armv8.x . Figures!

 

[scineram]

Doesn’t seem to have AMD level memory encryption.
Also interesting tidbit about ISA issues in Arm server adoption.

 

[Hitman928]

Also interesting tidbit about ISA issues in Arm server adoption.

Where is this mentioned?

 

[scineram]

Where is this mentioned?

Anandtech, page 1.

 

[Hitman928]

Anandtech, page 1.

Maybe be a bit more specific? I read the article but didn't see where ISA issues were mentioned.

 

[scineram]

Generally, I see SVE2 as probably the most important factor that would warrant the jump to a v9 nomenclature as it’s a more definitive ISA feature that differentiates it from v8 CPUs in every-day usage, and that would warrant the software ecosystem to go and actually diverge from the existing v8 stack. That’s actually become quite a problem for Arm in the server space as the software ecosystem is still baselining software packages on v8.0, which unfortunately is missing the all-important v8.1 Large System Extensions.

 

[Hitman928]

Ok, so by issues you just mean being stuck on the 8.0 baseline. I thought you were referring to some kind of issues with the ISA in the server space. Thanks for clarifying.

 

[soresu]

So SVE2 really isn't a part of Armv8.x . Figures!

I assumed as much when it was announced 2 years ago.

The messaging they used for SVE2 and TME in that blog post implied multi year next gen R&D investments which I took to mean major ISA version change at the time.

At least NEON support will remain so that v8 SIMD code will not all have to be rewritten for v9 CPU cores, at least for ARM Ltd IP.

The lack of TME being mentioned in this announcement sounds like this won't make it into all v9-A IP cores as standard though.

Possibly just for Neoverse IP.

 

[Doug S]

The ARMv9 stuff says that "SVE2v2" is included. I thought SVE2 was an optional extension to ARMv8, one which apparently no one ever included (though one would expect it was to be included in Nuvia's design)

 

[soresu]

I thought SVE2 was an optional extension to ARMv8, one which apparently no one ever included

It was never outlined in the initial 2019 announcement exactly what the strategy was for SVE2, beyond the fact that it was a multi year development, and that NEON would still be supported alongside SVE2 implementations for legacy code.

Hence the speculation over it back then about it being part of ARMv9 as standard.

Which I called I might add.

My smugness at being right will outlast the heat death of the universe!

 

[Doug S]

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.

 

[Thala]

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.

Technically Apples cores are ARMv8.6 so they will appreciate that many of the optional features are merged into ARMv9. A agree, ARMv9 would have been a nice baseline for ARM Macs.

Interestingly when programming for Windows i always assume ARMv8.2 baseline without explicitly checking the hardware. I guess in the future i need to check HW capabilities...

 

[DrMrLordX]

I doubt that MS's OS decisions will make Apple change their supported ISA at all. Mac hardware doesn't have to run Windows - that's pretty niche (Boot Camp).

In any case, I expect that if NV actually buys out ARM, that Apple will stick with their current license and diverge from future ARM development. Unless NV makes future licenses very attractive to Apple.

 

[jpiniero]

I'd say it'd be in Apple's best interest to minimize the downside of staying on ARM ISA if the nVidia deal goes through.

 

[Roland00Address]

About time! Been waiting for so long. In 2019 we were speculating a 2020 announcement. Perhaps that was wrong expectations, covid, or it took longer than expected but I am happy it is finally here!

( goes to read , now )

 

[Thala]

In any case, I expect that if NV actually buys out ARM, that Apple will stick with their current license and diverge from future ARM development. Unless NV makes future licenses very attractive to Apple.

Thats like Apple shooting themselfs in the foot. They benefit massively from the ARM ecosystem. Thats also the reason any change Apple ever wanted was always going into an ARMv8.x version - and this also the reason tat Apple Cores are ARMv8.6 and ARMs own cores are ARMv8.2.
Their only chance to succeed with an ISA switch was always tightly related to the vast ARM ecosystem.

I doubt that MS's OS decisions will make Apple change their supported ISA at all. Mac hardware doesn't have to run Windows - that's pretty niche (Boot Camp).

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).

 

[jpiniero]

Thats like Apple shooting themselfs in the foot. They benefit massively from the ARM ecosystem. Thats also the reason any change Apple wanted was always going into an ARMv8.x version - and this also the reason tat Apple Cores are ARMv8.6 and ARMs own cores are ARMv8.2.

nVidia buying ARM changes the whole equation. I don't know what works best for Apple but a hard fork seems all but inevitable if switching ISA to something not ARM is too much. Course we don't know if it will actually go through.

 

[Thala]

nVidia buying ARM changes the whole equation. I don't know what works best for Apple but a hard fork seems all but inevitable if switching ISA to something not ARM is too much. Course we don't know if it will actually go through.

A different company owning ARM does not change much for Apple at all - they will obtain an architecture license and go on. Diverging from the ecosystem would change everything.

 

[soresu]

and this also the reason tat Apple Cores are ARMv8.6 and ARMs own cores are ARMv8.2

This is badly messaged on ARM's part, and it makes them appear rather slipshod on the PR side of things I think.

Apparently despite advertising only ARMv8.2-A even up to A78/X1 they do in fact support a limited subset of features and instructions from several more recent ISA increments.

 

[jpiniero]

they will obtain an architecture license and go on.

That's what I mean, they won't want to do that with nVidia. Perhaps the question is whether they can get a v9 license before the acquisition goes through and the terms are such that nVidia can't mess with it. Or whether tApple even wants to.

 

[Doug S]

Thats like Apple shooting themselfs in the foot. They benefit massively from the ARM ecosystem. Thats also the reason any change Apple ever wanted was always going into an ARMv8.x version - and this also the reason tat Apple Cores are ARMv8.6 and ARMs own cores are ARMv8.2.
Their only chance to succeed with an ISA switch was always tightly related to the vast ARM ecosystem.

Diverging wouldn't mean abandoning ARMv8, it would mean not following to ARMv9. Android hasn't even dropped AArch32 yet, it will be at least a decade before it could conceivably go ARMv9 only.

Given that there is no "vast" (more like near zero) desktop/laptop ARM ecosystem, that clearly wasn't a factor in the Mac's ISA switch - it was their OWN iOS-based ARM ecosystem that paved the way for a successful ISA switch. Had they been on some Apple proprietary ISA for iOS instead of ARM, a switch to ARM for the Mac would have been an utter failure - but they could have switched the Mac to that Apple only ISA quite successfully. Their only loss in doing so versus the current situation with ARM would have been the lack of a native Windows port.

Apple benefited greatly from the ARM ecosystem early on when they were dependent on it for stuff like Samsung designed SoCs and drivers, if they did a proprietary ISA or something that had no infrastructure like RISC-V they could never have pulled it off - at least not in 2007 since they would have had to build so much more themselves. However, they passed the point where the non-Apple ARM world impacts their fortunes about a decade ago, they are a part of it but function independent of it. Those untold billions of embedded ARM CPUs in microcontrollers, modems, routers, autos et al don't benefit Apple at all anymore. If that entire embedded ARM market along with Android switched en masse to RISC-V tomorrow, Apple would barely notice or care.

 

[Thala]

Apple benefited greatly from the ARM ecosystem early on when they were dependent on it for stuff like Samsung designed SoCs and drivers, if they did a proprietary ISA or something that had no infrastructure like RISC-V they could never have pulled it off - at least not in 2007 since they would have had to build so much more themselves. However, they passed the point where the non-Apple ARM world impacts their fortunes about a decade ago, they are a part of it but function independent of it.

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.

Look where RISC-V stands today - years after its introduction. This should give you a warning example of how hard it is without existing ecosystem. Heck, even most build systems used to even compile above mentioned frameworks are lacking a cross-compile options for RISC-V. But there are more problems. UEFI/ACPI are essentially only defined for ARM and x86 today - have fun build standardized HW without these interface specifications.

Also the fact that UEFI/ACPI is standardized for ARM (besides ARMv8 ISA support of course) is the main reason, that you can just boot generic Windows images without any specific modifications on your M1 Mac in a VM.

That having said, i do not know if Apple jumps on the ARMv9 bandwagon this year or maybe 2 years down the road. Until then they will stay with ARMv8.x.