" It's an open secret that ARM was (and still is) selling the CPU design at a discount if you integrated their Mali (GPU)."
Why is that bad? Not only it's common business practice (the more you buy from us, the cheaper we sell), it also makes sense from the support perspective. Support the integration between their cores and a different GPU would be more work for them than integration of their cores with their own GPUs.
That's why companies expand to adjacent markets: efficiency.
A completely different thing would be to say: "if you want our latest AXX core, you have to buy our latest Mali GPU". That's bundling, and that's illegal.
No. Precisely the Tegra SoC within the Nintendo Switch (X1) uses ARM Cores. Specifically A57 and A53. NVIDIA's project to develop their own v8.2 ARM-based chip is called Denver.
What do you define the difference between an in-house developed SoC with ARM-IP Cores and Denver's "ARM-based chip"? It is going to be a new architecture, but using a mix of ARM IP blocks and in-house IP following the ARM ISA?
"and may just fork off on the ARMv8.3 spec, adding a few instructions here or there"
No, they may not. People keep suggesting these kinds of things, but part of the license agreement is that you can't modify the ISA. Only ARM can do that.
Well, regardless of whether this amendment is kosher or not, AMX definitely exists. Perhaps the $2T tech behemoth was able to work out a sweetheart deal with the $40B semiconductor company.
> There’s been a lot of confusion as to what this means, as until now it hadn’t been widely known that Arm architecture licensees were allowed to extend their ISA with custom instructions. We weren’t able to get any confirmation from either Apple or Arm on the matter, but one thing that is clear is that Apple isn’t publicly exposing these new instructions to developers, and they’re not included in Apple’s public compilers. We do know, however, that Apple internally does have compilers available for it, and libraries such as the Acclerate.framework seem to be able to take advantage of AMX. [0]
my123's instruction names leads to a very shallow rabbit hole on google, which turns up a similar list [1]
Now custom instructions are directly on the regular instruction space...
(+ there's the can of worms of target-specific MSRs being writable from user-space, Apple does this as part of APRR to flip the JIT region from RW- to R-X and vice-versa without going through a trip to the kernel. That also has the advantage that the state is modifiable per-thread)
That's like saying that my Intel CPU comes with an NVIDA Turing AI acceleration extension. The instructions the CPU can run on an Apple ARM-based CPU is all ARM ISA. That's in the license arrangement, if you fail to pass ARM's compliance tests (which include not adding your own instructions, or modifying the ones included) you can't use ARM's license.
Please, stop spreading nonsense. All of this is public knowledge.
No. I reverse-engineered it and AMX on the Apple A13 is an instruction set extension running on the main CPU core.
The Neural Engine is a completely separate hardware block, and you have good reasons to have such an extension available on the CPU directly, to reduce latency for short-running tasks.
Is it possible AMX is implemented with the implementation-defined system registers and aliases of SYS/SYSL in the encoding space reserved for implementation-defined system instructions? Do you have the encodings for the AMX instructions?
The AMX is an accelerator block... If you concluded otherwise, your reverse-engineering skills are not great...
Let me repeat this: part of the ARM architectural license says that you can't modify the ISA. You have to implement a whole subset (the manual says what's mandatory and what's optional), and only that. This is, as I've been saying, publicknowledge. This is how it works. And there are very good reasons for this, like avoiding fragmentation and losing control of their own ISA.
After your tone, not certainly obligated to answer but will write one quickly...
Apple A13 adds AMX, a set of (mostly) AI acceleration instructions that are also useful for matrix math in general. The AMX configuration happens at the level of the AMX_CONFIG_EL1/EL12/EL2/EL21 registers, with AMX_STATE_T_EL1 and AMX_CONTEXT_EL1 being also present.
The main source of revenue for ARM is, by far, royalties. Licenses are paid once, royalties are paid by unit shipped. And they shipped billions last year.
Revenue is not $300, we don't know what ARM's revenue is because it hasn't been published since 2016. And back then it was like $1.5 billion. $300 million was net income. Again, in 2016.
I think you've already been adequately corrected on your misconceptions about ARM's CPU design teams.
Do you think a multi-billion corporation would pay a low level employee an extra million just because they can? You may overestimate the value of individual engineers, big corporations don't. That's how they become big. The money they make is the difference between the value of your work and the money you make. The smartest and best connected people in the company are working to make that gap as wide as possible.
Yes, but if this engineer's achievements are well known, competition between companies is what drives crazy comp because they are seen more like a strategic asset rather than another engineer. They wouldn't quite be a low-level employee, these unicorn ICs often report directly to middle management or above. I'm surprised how little industry experience some of these commenters seem to have given that it's HN.
Competitors are also multi-billion corporations. I do work for one of those big multi-billion corporations, it's my second, and I've also worked for others not so big. Only newbie engineers fresh out of college believe the myth of the genius engineer who gets grossly overcompensated because they're so smart. You can find 10s of thousands of brilliant engineers in any of these companies, that's the bottom line. Not a single one of them is, on their own, irreplaceable. Specially valuable individuals get awarded distinctions like "fellow" or "distinguished". They're valuable, more than anything else, because of their contacts and rapport within the industry. It's never technical competency; not that they're technically incompetent, but most of their underlings will likely be more technically competent than them (if they're smart, after all, they will do the technical work). If you haven't figure this out yet, don't worry, you'll get there.
You clearly don't work for one of the ones that grant 7 figure salaries. I'm not saying they're common or that any engineer can aspire to achieve one, I'm saying they exist, I've seen it first hand, and I don't understand why it's so hard for you to accept they exist. Nothing in your comment constitutes novel insight to me, and neither of us have a good measure as to which of us have more credibility than the other, but I suspect you're just judging based on your own narrow experience.
Edit, just for more context I'm speaking of FAANG level companies here and very rare individual unicorn engineers who have been specifically hired into these kinds of positions for past achievements that have impact across the whole industry. I would agree with your general skepticism in any other context.
As I said, I'm on my second FAANG. The "very rare individuals" you mention are hired L9 or above. That is, distinguished engineers+. You don't get to L9 with "a valuable technical contribution", you get there because people know who you are, you have strong network of connections within the industry, and you are in a position to make strategic decisions. It's very much not a technical position, it's borderline executive. Let's put it this way, the people with that kind of compensation, you know who they are. It's never an anonymous whizz kid who's very good at solving technical problems, it's the guy who hired them and/or knows how to direct their work.
As you said, you don't know me and I don't know you, so I don't have a reason to doubt your word. If you say you've met engineers who get that kind of compensation, I'll give you the benefit of the doubt. Everyone I've met or I've known to be in that level of compensation were the people I already knew were making that kind of money.
ARM was publicly traded between 1998 and 2016. In that period its value multiplied about 25x, not counting the premium of the acquisition. Could you elaborate, please? Where do you see the disaster? (Honest question).
Apple is a small, although significant, part of ARM's total market share. And that 25x is, as I said, without taking into account the premium. If you do, and there are good arguments to do so, the valuation growth is 35x, in almost 20 years.
Regarding innovation, ARM's been at it since 1990. I'm sure it's not the same now as it was 30 years ago, but we're well past the point where one can reasonably fear it to be an unsustainable business. Last time I heard numbers, they were talking about more than 50 billion devices shipped with ARM IP in them. That is a massive market.
You don't answer my question. Why wouldn't licensing businesses work as publicly traded companies? What's the fundamental difference, specially in an increasingly fabless market, between a company licensing IP to other companies and a company selling productized IP to consumers?
Definitely in the short run, because of the understandable fear from NVIDIA's competitors to use their (now) technology. Maybe in the mid run if those fears begin to crystallize. Unlikely in the long run, I'd assume NVIDIA would spin ARM off before killing it entirely, buying ARM would be a multi-billion investment.
They can also design their own ISA. An ISA is a document, they can write their own. Now, can you think of reasons why they wouldn't want to do it that don't also apply to MIPS?
Why is that bad? Not only it's common business practice (the more you buy from us, the cheaper we sell), it also makes sense from the support perspective. Support the integration between their cores and a different GPU would be more work for them than integration of their cores with their own GPUs.
That's why companies expand to adjacent markets: efficiency.
A completely different thing would be to say: "if you want our latest AXX core, you have to buy our latest Mali GPU". That's bundling, and that's illegal.