How do we know ARM won’t be like PowerPC?

[LinkRS]

Aside from what others said, these two aren’t comparable. ARM has been around for decades and making an in-house CPU+GPU means you can optimize the software to it.

Look how fast and smooth iPadOS is. This is what macOS should feel like. Long term Mac users have always been annoyed the macOS support for third party graphics chips hasn’t been as good as Windows/Linux. With Apple using their own GPU and decoder/encoded chips they no longer have to lean on third party documentation and or have high power consumption causing throttling.

Yes, this. The biggest advantage with Apple moving to in-house hardware is control and optimization. This is also the reason that Mac computers have always felt more "together" than Windows or even UNIX machines. Apple developed (designed) the hardware and software to work together, which is not something that gets done (typically) in the Windows world. With an Apple designed SOC, where they control all aspects of it, they can optimize and fine tune both the physical chip, and the OS (software). If they need something to play back 4K video, they can add a custom block onto their SOC to handle the decoding, freeing up the main CPU from the task. They could then add a custom encoder and update Final Cut Pro to use that. To the end-user, they get a seamless fluid experience that would just not be possible on a generic CPU alone. The performance for AppleSI Macs is going to be so much more than just the raw performance from the ARM core. This is why benchmarks like GeekBench are worthless for something like AppleSI. The performance for Macs will be a sum of all of its parts, not just an ARM core. That is why this architecture shift is exciting! It is going to allow Apple to do what made them "Insanely Great" in the first place, be creative and create new experiences.

 

[springsup]

Apple isn’t abandoning Intel for ARM.

Apple are abandoning Intel for Apple Silicon.

The biggest difference isn’t the technology - it’s who is making the chips, who is responsible for the roadmap, etc. These are bespoke chips that are specifically designed for whatever Apple wants.

If there is any particular operation that happens to come up a lot for Apple’s customers, they can sink it down to the hardware for a level of performance off-the-shelf parts can’t match. Take the Afterburner cards in the Mac Pro - a future ASi chip tuned for pro video could integrate some or all of that technology in the main chip itself, rather than being an add-on card. Apple’s OS uses a lot of machine learning, and now they can bring over accelerator chips like the neural engine. There are other kinds of accelerators that could be built - for text processing, compression, networking, web browsing, etc. A world of opportunities has opened.

 

[Arctic Moose]

*Points towards light bulb*

You do realize Edison bulbs have been outlawed in Europe for years due to their inefficiency, and LED technology is evolving rapidly? (The first low-energy bulbs I purchased are obsolete now, non-dimmable, low brightness and high energy consumption.)

Speaking of sockets, non-DCL types are not allowed for new installations, so the round type will be obsolete in a few years:

Museum of Plugs and Sockets: lamp sockets and plugs

Annotated display of various types of celing lamp sockets and plugs

www.plugsocketmuseum.nl

 

[boss.king]

*Points towards light bulb*

A light bulb is a concept that can be executed in a number of ways. Incandescent bulbs are obsolete, then fluorescent bulbs, and now LED bulbs are becoming the norm in many places.

 

[casperes1996]

You do realize Edison bulbs have been outlawed in Europe for years due to their inefficiency, and LED technology is evolving rapidly? (The first low-energy bulbs I purchased are obsolete now, non-dimmable, low brightness and high energy consumption.)

Speaking of sockets, non-DCL types are not allowed for new installations, so the round type will be obsolete in a few years:

Museum of Plugs and Sockets: lamp sockets and plugs

Annotated display of various types of celing lamp sockets and plugs

www.plugsocketmuseum.nl

A light bulb is a concept that can be executed in a number of ways. Incandescent bulbs are obsolete, then fluorescent bulbs, and now LED bulbs are becoming the norm in many places.

I did also point out I was just joking around; There are LED bulbs in the lights above me actually

 

[Joelist]

Keep in mind that Apple Silicon (not ARM) is coming from a unit that Apple built from the ground up, and a lot of that build was raiding Intel, Qualcomm and others for top talent (like Johnny Srouji). Apple Silicon actually started after the A Series because up to A5 they were Cortex - starting at A6 they went to a completely different microarchitecture. And A6 was also when both per core performance and PPW improved exponentially.

There are other threads in here that link to some articles by people describing in general terms the new microarchitecture - short fat pipes with super accurate branch prediction. It was also described as desktop class coming all the way from A6. What is also interesting is that what it resembles (if anything) is actually Core 2.

 

[Yebubbleman]

PowerPC processors were a RISC-based processor that everyone claimed was an advanced design, superior to Intel, that could outperform Intel even at lower clock speeds. But as we all know, within a few years on the market, x86 chips kinda... kicked their ***.

Now Apple is switching once again to a RISC architecture that they claim is superior, at least on paper.

What went wrong with PowerPC? How do we know it won’t happen with ARM?

PowerPC didn't have a roadmap that emphasized performance per watt. Intel did for a while, but circa 2015, they began to have problems meeting their roadmap goals. Part of this is technological hurdles and part of it is lousy upper-management. But, suffice it to say that, barring the increase in core count in 8th Gen Intel, there hasn't been any substantial breakthrough in performance on Intel CPUs since 2015's Skylake.

As for ARM, Apple isn't just switching to ARM, it's basically making its own ARM architecture chips, hence "Apple Silicon". With PowerPC, Apple was reliant on Freescale (formerly Motorola) on the G4 and IBM on the G3 and G5 to make breakthroughs. With Intel, Apple was reliant on...well...Intel. Now, Apple is using their own chips and is, effectively in command of the ship. All ARM does is create general blueprints for the architecture that Apple can license and do whatever it wants with.

That's WAY more control than Apple had with either of the two previous architectures. Does that negate the possibility that Apple will hit a similar performance per watt wall with their own ARM based silicon? Not at all. But it won't happen for a long while (at least the same amount of time that we've been on Intel, if not way longer) and since they are fully in control of the processor roadmap, they'll be able to make any subsequent architecture switch all the more graceful when and if that time comes.

 

[MisterMe]

...

That's WAY more control than Apple had with either of the two previous architectures. Does that negate the possibility that Apple will hit a similar performance per watt wall with their own ARM based silicon? Not at all. But it won't happen for a long while (at least the same amount of time that we've been on Intel, if not way longer) and since they are fully in control of the processor roadmap, they'll be able to make any subsequent architecture switch all the more graceful when and if that time comes.

Point of clarification--Apple has used four (4) previous microprocessor architectures, three (3) of which powered the Macintosh. This are as follows:

1. MOS Technologies 6502--powered the Apple I, Apple ][, and Apple ///
2. Motorola 680x0--Lisa and Macintosh (1983-1994)
3. PowerPC--(1994-2006)
4. Intel x86--(2006-2020)

This list does not include the processors at the hearts of other Apple digital devices.

 

[MysticCow]

PowerPC Macs were sheer monsters for a long time and they could at lower clock speeds. Intel chips were horrendously inefficient and so had to ramp up the clock speed just to compete. That was back around the turn of the century...

Intel adapted and gave "faster, smaller, cooler" while PPC absolutely stagnated on "faster" and simply couldn't deliver. There could never be a G5 PowerBook because you'd likely need liquid cooling in a PowerBook...which is bad. The G4's couldn't handle DDR RAM at all. This is why the transfer to Intel happened--Intel could deliver better chips than anything the PowerPC alliance could ever give for the Mac.

With ASMacs, the potential exists for Apple to offer "faster, smaller, cooler" in ways Intel simply can't right now. Unless Apple royally screws up (and it could), the first ASMacs should show faster clock speeds (consumers understand bigger numbers = better performance) AND smaller chips AND chips that may only need iMac G3 levels of cooling in a laptop.

 

[Joelist]

If you want to understand Apple Silicon better go look at the history of Intel Israel - Apple poached a lot of those people for its own silicon design and development including Johnny Srouji. Intel Israel set the entire processor world on its ear when they designed the Banias microarchitecture. They also designed Dothan, Nehalem and Conroe (which is what the Apple Silicon microarchitecture resembles somewhat).

As a result, because Apple Silicon has been desktop class in architecture since A7 they can scale up easily. Right now A14 has 2 power and 4 efficiency cores - all they need to do is use the same core design just bump up the core counts to something like 6 power and 6 efficiency and scale the cache up correspondingly.

 

[Andropov]

With ASMacs, the potential exists for Apple to offer "faster, smaller, cooler" in ways Intel simply can't right now. Unless Apple royally screws up (and it could), the first ASMacs should show faster clock speeds (consumers understand bigger numbers = better performance) AND smaller chips AND chips that may only need iMac G3 levels of cooling in a laptop.

I feel like this was a much bigger thing in the 00s. Nowadays, with all the TurboBoost / dynamic clock speed complexity added on top, it's much harder to infer performance from the CPU frequency alone and people often turn to benchmarks anyway.

 

[progx]

There's no money in making console chips. It's the whole reason NVIDIA got out of them after the PS3.

While you're right about lack of profit in console chips, but you're wrong about NVIDIA.

NVIDIA is still in the console market, they're powering the Nintendo Switch with the Tegra. Before the Switch was released, NVIDIA touted their new relationship with Nintendo will last 20 years. Seeing how the Switch has done well, plus the Shield tablet has been discontinued, sounds like Nintendo picked the right company for their hardware. NVIDIA has all the entire system of the console, they're not sharing it with Intel (original Xbox) or IBM this time.

As for the future versions of the Switch, I've heard rumors it's a different NVIDIA SOC and APU they'll be running and not the Tegra and moving to "Volta." The Switch Pro isn't coming out, so who knows what NVIDIA and Nintendo are working on for the next Switch.

Nvidia Expects 20-Year Relationship With Nintendo Thanks to Switch

Nvidia CEO Jen-Hsun Huang calls the Switch an 'incredible game console.'

www.pcmag.com

Rumoured Switch Pro Might Not Actually Use Nvidia's Tegra X1+ Chip

And don't expect major performance improvements or 4K support

www.nintendolife.com

 

[Krevnik]

They were.

The reason PPC "fell behind" (for the markets Apple target, like laptops) is that IBM and Motorola, who were largely responsible for the PPC architecture were not interested in mobile as it was a tiny segment of the market. They continued (and still do) to build PPC for servers, which wouldn't fit in mobile computers like Apple wanted to build.

IBM wasn’t. Motorola was so interested in mobile they were bypassing laptops and focusing on embedded chips, including PowerPC derived designs at one point. Only to lose out to ARM-based designs in the end. Go figure.

I honestly don't believe that, it's just the history revisioning. "We could, we just didn't wanna" is loser talk to save face.

We shouldn't forget they did the Wii, XBOX and PS3 chips around this time frame too. IBM had a lot of consumer chips on the go but weren't able to deliver. Remember Steve Jobs promising us 3GHz G5's? - They told him they would be able to deliver that, never happened. That's a failure to execute not a decision not to provide it, he never would have said it if they hadn't promised it.

And they went all in on the custom Cell processor with Sony and Toshiba. They clearly had the will to do special projects. That chip was PowerPC based and could have superseded the G5 to be honest, sure the SPE design made it complicated to program for but it had potential in later iterations or even Apple could have joined that alliance and been involved in the development process to get a custom version with more PPE's which is exactly what Microsoft did when they wanted a processor for the XBOX 360 (IBM took the PPE design from Cell, tripled it up and gave it to Microsoft in a custom design).

IBM in my eyes for sure couldn't execute on a competitive product. It's also important to realise that the Conroe architecture from Intel was just way too good. It didn't just eclipse IBM's G5 and their PowerPC range entirely across all segments (including servers) it also decimated AMD and it has taken them just over a decade to recover.

The PPE wasn’t really a successor to the G5. They were PPC970 variants themselves, which included much more limited OoO capabilities compared to the PPC970 Apple used. The ask of customizing an existing design is different than asking for a truly new design.

I wish could find a source, but at the time, I was thinking that Toshiba’s contributions to Cell were the SPEs, but I could be wrong.

But honestly, with the market pushing towards laptops, the PPC970 variants that went into consoles wouldn’t have done much of anything for Apple. They needed a new design for laptops that simply wasn’t coming, not tweaked versions that were in ways side-grades of the chips they already had.

While you're right about lack of profit in console chips, but you're wrong about NVIDIA.

NVIDIA is still in the console market, they're powering the Nintendo Switch with the Tegra. Before the Switch was released, NVIDIA touted their new relationship with Nintendo will last 20 years. Seeing how the Switch has done well, plus the Shield tablet has been discontinued, sounds like Nintendo picked the right company for their hardware. NVIDIA has all the entire system of the console, they're not sharing it with Intel (original Xbox) or IBM this time.

As for the future versions of the Switch, I've heard rumors it's a different NVIDIA SOC and APU they'll be running and not the Tegra and moving to "Volta." The Switch Pro isn't coming out, so who knows what NVIDIA and Nintendo are working on for the next Switch.

I’m not sure what you mean by Volta replacing Tegra here. Volta is not an SoC, but a GPU architecture. The SoC mentioned in rumors was called Xavier (which includes a Volta derived GPU), which itself is still considered a Tegra SoC. One that’s been available for a couple years now.

The problem I see is that after Xavier, the SoCs that Nvidia have announced are chips aimed at robotics and autonomous vehicles. So Nintendo is going to be increasingly dependent on chips where they are the only customer (customized derivative designs), most likely. They might not have an issue with this, as it’s not new to them, they’ve used these sort of one-off PPC variants for the GameCube, Wii and WiiU. But I do wonder what would stop them from shopping around for another ARM provider. Nvidia has yet to do anything custom for Nintendo, and seem less interested in the console/set-top-box space these days compared to when they were trying to make the Shield work. I hope I’m wrong, but if I was Nintendo, I would be somewhat wary of Nvidia right now.

 

[Quu]

While you're right about lack of profit in console chips, but you're wrong about NVIDIA.

NVIDIA is still in the console market, they're powering the Nintendo Switch with the Tegra.

This is more nuanced though. What I said about NVIDIA getting out of console chips is still true for the major consoles. And it's not me saying that's the reason, it's the CEO of NVIDIA on record saying why they stopped bidding on console manufacturers part contracts.

The SoC in the switch is different because NVIDIA was already developing that SoC regardless of Nintendo for their own shield products. They didn't need to reinvent the wheel.

What Jensen Huang was saying when asked why we're not seeing NVIDIA chips in consoles (2013's PS4 and XBOX ONE, all prior to the 2017's Nintendo Switch) that was the answer he gave. He specifically said these chips take engineering time from their own graphics development and have very slim margins on each chip sold making it an undesirable market to them.

And now once again where is NVIDIA in the PS5 and the new XBOX Series X/S ? - Absent. Both consoles have gone with an SoC from AMD.

Now to be clear, I'm not suggesting NVIDIA isn't in the console market, obviously the switch uses an SoC from them but my point is they used to do custom designs (PS3's RSX GPU based loosely on a 7000 series GPU but had two re-spins for lower cost and power consumption they also did the original XBOX chip) but now they realised custom in consoles means no money. Reusing a chip they already made however, that's lucrative.

You can read more about this from 2013: https://www.extremetech.com/gaming/150892-nvidia-gave-amd-ps4-because-console-margins-are-terrible

 

[progx]

I’m not sure what you mean by Volta replacing Tegra here. Volta is not an SoC, but a GPU architecture. The SoC mentioned in rumors was called Xavier (which includes a Volta derived GPU), which itself is still considered a Tegra SoC. One that’s been available for a couple years now.

The problem I see is that after Xavier, the SoCs that Nvidia have announced are chips aimed at robotics and autonomous vehicles. So Nintendo is going to be increasingly dependent on chips where they are the only customer (customized derivative designs), most likely. They might not have an issue with this, as it’s not new to them, they’ve used these sort of one-off PPC variants for the GameCube, Wii and WiiU. But I do wonder what would stop them from shopping around for another ARM provider. Nvidia has yet to do anything custom for Nintendo, and seem less interested in the console/set-top-box space these days compared to when they were trying to make the Shield work. I hope I’m wrong, but if I was Nintendo, I would be somewhat wary of Nvidia right now.

These are all rumors. We won't know what Nintendo's new hardware will be until it is announced. Nintendo, much like Apple, is one of (if not THE) most secretive companies on the planet. There were rumors Nintendo was getting ready to pair up with someone else.

The current strong rumor is the "Switch 2" is new hardware. Apparently, it is being supplied by NVIDIA again, plus this time it's custom compared to the Tegra X1. While I don't know how this is an obstacle, but I don't know how NVIDIA builds their SOC, but it isn't compatible with current Switch software. "Volta" is an old rumor, plus it was heavily geared toward what would have been called "Switch Pro." Due to the supply conditions, the rumors Nintendo has abandoned it and is moving toward the new hardware. All I know is NVIDIA and Nintendo's partnership isn't going to change in the near future.

The only thing that has been constant in these stories is they are sticking with the current Switch approach and will not be going in a different direction with the design. My guess is Nintendo is going to be a gaming device company and no longer a console. Does this mean there won't be a TV-only version? No. They could just brand the Shield TV into the Switch family.

 

[metapunk2077fail]

The operating system can be ported to any processor. New architectures are always a decade or two away.

 

[JMacHack]

These are all rumors. We won't know what Nintendo's new hardware will be until it is announced. Nintendo, much like Apple, is one of (if not THE) most secretive companies on the planet. There were rumors Nintendo was getting ready to pair up with someone else.

The current strong rumor is the "Switch 2" is new hardware. Apparently, it is being supplied by NVIDIA again, plus this time it's custom compared to the Tegra X1. While I don't know how this is an obstacle, but I don't know how NVIDIA builds their SOC, but it isn't compatible with current Switch software. "Volta" is an old rumor, plus it was heavily geared toward what would have been called "Switch Pro." Due to the supply conditions, the rumors Nintendo has abandoned it and is moving toward the new hardware. All I know is NVIDIA and Nintendo's partnership isn't going to change in the near future.

The only thing that has been constant in these stories is they are sticking with the current Switch approach and will not be going in a different direction with the design. My guess is Nintendo is going to be a gaming device company and no longer a console. Does this mean there won't be a TV-only version? No. They could just brand the Shield TV into the Switch family.

Honestly I could see Nintendo partnering with AMD again if NVidia isn’t willing to play ball. The PPC chips used through the GC and Wii/WiiU era were part custom AMD designs.

The reason the Switch is NVidia in the first place is that it’s literally an NVidia shield at its core.

If AMD is being honest about gearing up ARM design again, a custom SoC might make a future reappearance in a future console. (Not likely Switch2).

This is all conjecture of course, but it’s all dependent on what NVidia is willing to do.

 

[Krevnik]

The current strong rumor is the "Switch 2" is new hardware. Apparently, it is being supplied by NVIDIA again, plus this time it's custom compared to the Tegra X1. While I don't know how this is an obstacle, but I don't know how NVIDIA builds their SOC, but it isn't compatible with current Switch software. "Volta" is an old rumor, plus it was heavily geared toward what would have been called "Switch Pro." Due to the supply conditions, the rumors Nintendo has abandoned it and is moving toward the new hardware. All I know is NVIDIA and Nintendo's partnership isn't going to change in the near future.

The only thing that has been constant in these stories is they are sticking with the current Switch approach and will not be going in a different direction with the design. My guess is Nintendo is going to be a gaming device company and no longer a console. Does this mean there won't be a TV-only version? No. They could just brand the Shield TV into the Switch family.

Then I’m not sure why you tried to tie the rumors to “Volta” if the rumors are that Xavier has been abandoned. Again, Volta isn’t an SoC, Tegra is. So the phrasing can be hard to parse when claiming an SoC product line is being replaced with a GPU architecture. And surely, anything Nintendo gets that’s custom is still going to be Tegra derived, as Nvidia is still producing new Tegra SoCs, with Atlan being announced this year which uses a Ampere-based GPU design.

And again, I didn’t discount that Nvidia would be willing to do something one-off for Nintendo. Just more stating that if I were Nintendo, I’d be expecting some sort of assurances that Nvidia still cares about the space, because at least publicly Nvidia has stated that it intends to focus it’s SoCs into automation going forward. So GPUs with more compute/ML potential compared to rasterization. In other words, their primary efforts aren’t in great alignment with Nintendo’s at the moment. There’s still some overlap, but considering Nintendo has been perfectly willing to dump partners with little warning in the past, I wouldn’t put much emphasis on something Nvidia puts out about their relationship.

 

[progx]

Then I’m not sure why you tried to tie the rumors to “Volta” if the rumors are that Xavier has been abandoned. Again, Volta isn’t an SoC, Tegra is. So the phrasing can be hard to parse when claiming an SoC product line is being replaced with a GPU architecture. And surely, anything Nintendo gets that’s custom is still going to be Tegra derived, as Nvidia is still producing new Tegra SoCs, with Atlan being announced this year which uses a Ampere-based GPU design.

And again, I didn’t discount that Nvidia would be willing to do something one-off for Nintendo. Just more stating that if I were Nintendo, I’d be expecting some sort of assurances that Nvidia still cares about the space, because at least publicly Nvidia has stated that it intends to focus it’s SoCs into automation going forward. So GPUs with more compute/ML potential compared to rasterization. In other words, their primary efforts aren’t in great alignment with Nintendo’s at the moment. There’s still some overlap, but considering Nintendo has been perfectly willing to dump partners with little warning in the past, I wouldn’t put much emphasis on something Nvidia puts out about their relationship.

First off, yes "Volta" is a GPU, but in the context of the article I was reading is incorrect. I haven't talked about this stuff in quite some time, so I completely forgot about Xavier. From YouTube videos to posts, random news articles, who knows what is in the next Switch. It's safe to assume it'll be some from of Arm based SOC.

Secondly, yes, I know Nvidia wants to move more into automation and industrial. What does this mean for Nintendo and their expectations? I don't know. However, IBM and AMD never touted their relationship in a press release with Nintendo either. If anything, Nintendo is bending to Nvidia's pipeline and not creating anything custom with them. Nintendo poured a lot of R&D into PowerPC and OpenGL, which they used for over a decade with GC, Wii and Wii U. Does it mean they're squarely sitting with them? No. Heard rumblings they were looking at Samsung and LG, but they could be testing out many different SOCs for future products or if the relationship fails. Which what company doesn't do that. I would imagine Apple is exploring RISC-V, in case Arm gets bought by Nvidia or another company.

 

[jdb8167]

I would imagine Apple is exploring RISC-V, in case Arm gets bought by Nvidia or another company.

I wouldn’t be surprised if Apple has the Xnu/Darwin kernel running on RISC-V but it seems unlikely that they are worried about Arm getting bought. They have an Arm architecture license that is probably licensed in perpetuity given the history of Apple and Arm. Apple isn’t tied to future Arm architecture changes.

 

[JMacHack]

First off, yes "Volta" is a GPU, but in the context of the article I was reading is incorrect. I haven't talked about this stuff in quite some time, so I completely forgot about Xavier. From YouTube videos to posts, random news articles, who knows what is in the next Switch. It's safe to assume it'll be some from of Arm based SOC.

Secondly, yes, I know Nvidia wants to move more into automation and industrial. What does this mean for Nintendo and their expectations? I don't know. However, IBM and AMD never touted their relationship in a press release with Nintendo either. If anything, Nintendo is bending to Nvidia's pipeline and not creating anything custom with them. Nintendo poured a lot of R&D into PowerPC and OpenGL, which they used for over a decade with GC, Wii and Wii U. Does it mean they're squarely sitting with them? No. Heard rumblings they were looking at Samsung and LG, but they could be testing out many different SOCs for future products or if the relationship fails. Which what company doesn't do that. I would imagine Apple is exploring RISC-V, in case Arm gets bought by Nvidia or another company.

I wouldn’t be surprised if Apple has the Xnu/Darwin kernel running on RISC-V but it seems unlikely that they are worried about Arm getting bought. They have an Arm architecture license that is probably licensed in perpetuity given the history of Apple and Arm. Apple isn’t tied to future Arm architecture changes.

Iirc Apple does have a perpetual license to ARM because they helped create it, and there was a rumor that they also investigate RISC-V.

 

[satcomer]

I want the return on YellowDog Linux as another bootcamp partition! I would pay a small fee to see that happen again!

 

[mr_roboto]

I wouldn’t be surprised if Apple has the Xnu/Darwin kernel running on RISC-V but it seems unlikely that they are worried about Arm getting bought. They have an Arm architecture license that is probably licensed in perpetuity given the history of Apple and Arm. Apple isn’t tied to future Arm architecture changes.

On the contrary, it is very likely they're worried about where Arm lands.

Yes, it's probably true that by the letter of the contracts they don't have to follow along.

Yes, they could flip to RISC-V.

But here's the problem with RISC-V: Apple is quite committed to Arm. I guarantee you they don't want to do yet another CPU ISA transition just for political reasons, especially not so soon after they just transitioned the Mac. No matter how smooth they make these transitions look, understand that it's an amount of work and disruption which makes it a once-a-decade kind of thing. The rewards have to outweigh the costs, and the costs are substantial.

As for the idea of going their own way with a private branch of Arm? They don't want that either. Apple benefits a lot from being a part of the community of companies implementing and using the Arm architecture. For example, there's tons of non-Apple contributors to the arm64 back end of LLVM/Clang.

By some insider accounts I've read, AArch64 could almost be called the "Apple ISA", because Apple collaborated with Arm Holdings to design it and footed a lot of the bills. The reason they let Arm Holdings wholly own the spec and license it to others is because it's not good (in this part of the tech industry, anyways) to try to be your own private island.

This is why Apple is never going to document their custom Arm ISA extensions like AMX for use outside Apple-provided libraries, or provide the internal tooling they use when writing libraries for same to the public. Not only is it probably against the terms of their architectural license (it's not necessarily a "you get to do anything you like" kind of deal), a company in Apple's position should not want to create a public-facing Apple-only Arm dialect. The only way such extensions will ever be usable by outsiders is if Apple decides there's a reason to contribute them back to Arm for incorporation into a future revision of the official spec.

And that's why Apple has to be worried about the prospect of Arm Holdings ending up being controlled by Nvidia. They (and other Arm licensees) need Arm Holdings to be a trusted partner.