Nope, that would take congressional approval and congressional leaders are all bought by the people that paid for the Iran Sanctions Act of 1996. At this point only DoD and CIA can make it happen, thus why I mention any of this.
Yah my Claude Code agents run a ton of Python and bash scripts. You're probably missing out on a lot of tool use cases without full tool use through POSIX compatibility.
A lot of government would be improved by making elected positions be very specific roles.
Why are we deciding military strategy from a guy that was elected to fix labor rights? Should the same guy running the school system also be in charge of selecting Supreme Court justices?
Also, the founding fathers had it right: An independent electoral college should decide elected positions, not the general public. Hiring decisions should be left up to people that are expert at hiring, not random people.
The only role the general public should have in government is deciding their representative - it's literally in the name!
And executive branch isn't supposed to be a representative. It's only role is to execute laws created by the representatives.
> he vast majority of people trust the Iranian government
Right... Nobody sane would trust an authoritarian regime which suppresses any type of free speech and and even banned the internet regardless of everything else.
Mistrusting Israel is understandable but that seems tangential.
I don't think I have the skillset and personality that would me allow to rise to the top of the hierarchy in a brutal totalitarian regime built on religious fanaticism. So no I would not do that.
> armed insurrectionists
Unfortunately not even remotely armed enough to make a difference...
> War changes rules of a country.
Oh so the Iranian regime was not murdering its own peacefully protesting citizens (regardless of the existence of these "armed insurrectionists") for many years now?
>I don't think I have the skillset and personality that would me allow to rise to the top of the hierarchy in a brutal totalitarian regime built on religious fanaticism.
This is about whether you would shut down the internet or not, not whether you would rise to the top of the hierarchy on a brutal totalitarian regime built on religious fanaticism... like Israel. You know, a country with strict limits on media, including shutting down media outlets it deems critical of the state.
Yes, you would shut down the internet in a war. Yes, specifically you. Just like how you would just down media companies and plane flights in wartime, since you, yes you specifically, do not believe in Democracy.
>Oh so the Iranian regime was not murdering its own peacefully protesting citizens (regardless of the existence of these "armed insurrectionists") for many years now?
So then for how many years do you think Mossad armed the insurrectionists that you are trying to call "peaceful protesters"?
"Brutal apartheid state"? Well perhaps... certainly not a totalitarian regime, though.
> many years do you think Mossad armed the insurrectionists
Sadly and unfortunately either not long enough and/or didn't provide them with enough weapons. But yeah I agree with your sentiment that Mossad should have done a much better job if they were serious about overthrowing the regime.
Great. Glad you agree that it was Mossad that was responsible for all the civilian deaths during the violent Iranian insurrection, and not the Iranian government.
I think we can all agree that the Iranian government are the good guys and the Israeli regime are the bad guys. That's not in dispute. What IS in dispute is how we work together on removing the Israeli government. I think we should support the Iranian government, since they are already well on their in the process of removing the Israeli regime from power and replacing them with the good Hamas government.
> It’s a comparison with Israel rather than a stand alone statement
Yes, I understood that and still it makes no sense to me, I mean extremely untrustworthy and very untrustworthy seems about the same since you can't trust anything either source says.
Israel at least have a free(ish) media and is less likely to hang someone leaking information from a construction crane.
This completely insane and sounds like Iranian propaganda. You can hate Israel but spreading propaganda is really terrible. You find it hard to believe that the regime that hangs women from a crane in the square for not wearing the hijab would not do this? There are videos of IRGC shooting into crowds and apartments but your view is just “Israel Bad”. Be seious. You can be against the war and still live in reality with everyone else.
That is not an argument, it is just weaseling. What did I make up? The great governor of California is a democrat and part of the Getty fortune. Perhaps you will believe them?
I honestly do see Trump just declaring he lost. "These Iranians put up a good fight and we weren't prepared." or something to that effect. He's been known to acknowledge defeat, like his complete 180 when Mamdani won.
Can we set up a bet on Polymarket or something? I'll take that action any day. He would have to phrase such an acknowledgement as an apology, and a Trump doesn't do that. Ever.
He didn't directly acknowledge defeat with Mamdani, at least not that I heard. It was more a case of populist game recognizing populist game.
It's done in a military court of a far-right theocratic government that doesn't believe in a Democracy, and is in fact attacking other Democracies like Lebanon and Iran. They don't even bother having fair trials. So it's basically kidnap Palestinian children as hostages, then kill them.
I still have some of the 88000 reference manuals, and it was really my first introduction to RISC architecture, and I thought it was great. But I never figured out why companies like Apple never chose it for their CPU?
I believe it was the first RISC that Apple prototyped building a Mac around, including a 68K emulator. IIRC from Gary Davidian’s CHM oral history, it was corporate dealmaking that led to AIM and PPC more than any technical negatives for the 88K.
Yeah, this is probably closest to the right answer. Apple DID choose the 88K, and then changed. Reportedly they put some 88K systems in a Mac chassis.
I do wonder what the exact reasons were. Maybe the PPC (complete systems) could be made cheaper? Maybe Apple was worried about relying on a single vendor? I am kind of skeptical of the “corporate dealmaking” angle, because it seems like there are valid technical reasons to NOT choose the 88K. Namely, that it requires companion chips, and the whole system (board + chips) ends up being complicated and expensive.
What I always read was that Apple did not want to be stuck relying only on Motorola again like they were with the 680x0. And it worked out, kinda, Apple had IBM to rely on to make the G5 (until IBM also lost interest)
I remember reading that the successor 88110 design with the support chips integrated was announced mainly to woo Apple but I don't know how true that is.
Bitsavers have some documents about the Jaguar RISC project[1] that do indicate Apple's feedback went into the 88110, for example in the System ERS it states "The main processor for the Jaguar is a new version of the Motorola 88000 family which has been enhanced (with input from Jaguar's team) in several areas over the existing implementation. This processor (which will be the MC88110) will be referred to as XJS in the ERS.". There's also an architecture document describing changes Apple wants to make to the 88000 ISA, although I'm not sure how much of this actually got through into the final 88110 (Apple wanted to break binary compatibility, not sure if that happened).
[1] The high end RISC machine project that went nowhere, which AFAIK became known as Tesseract when switched to PPC before it fizzled out.
Timing. The 68k still had legs, i.e. the 68040 provided great drop in performance and had an enormous ecosystem and economies of scale. By the time the RISC wars were starting to get fever pitched, the POWER architecture and AIM alliance seemed like a blessing to combine ecosystems and economies of scale for the A and M constituents. And it was.. successful product lines for 2-3 decades from all sorts of embedded systems to G5 workstations to spacecraft.
They did basically. What happened is that Apple own CPU project crashed and burned. Then they had some meetings with people including DEC for Alpha and IBM. IBM offered POWER and IBM was also willing to go in on some other projects, like the next gen OS Teligent.
But Apple didn't want to drop Motorola fully. So Motorola, Apple and IBM figured out that with some tweaks to the 88000 the could turn it into something POWER like. And that thing was PowerPC that Motorola supplied to Apple. That's my understanding.
Both Apple and NeXT had machines prototyped around it, but it was initially very expensive I believe, and I think Apple was easily convinced to go with PowerPC ... and rather than evolve it and push it further Motorola dropped it in favour of going in on PowerPC.
The sad thing is Intel showed there was still life left in CISC, and Motorola themselves ended up circling back on 68k in the form of ColdFire which proved you could do for 68k what Intel did w/ the Pentium. But by then all their 68k customers had moved on from the 68k ISA.
Motorola had made a few design mistakes, like adding memory indirect addressing in MC68020, which were removed much later, in the ColdFire Motorola CPUs.
But Intel had made much more design mistakes in the x86 ISA.
The truth is that the success of Intel and the failure of Motorola had absolutely nothing to do with the technical advantages or disadvantages of their CPU architectures.
Intel won and Motorola failed simply because IBM had chosen the Intel 8088 for the IBM PC.
Being chosen by IBM was partly due to luck and partly due to a bad commercial strategy of Motorola, which had chosen to develop in parallel 2 incompatible CPU architectures, MC68000 intended for the high end of the market and MC6809 for the low end of the market.
Perhaps more due to luck than due to wise planning, Intel had chosen to not divert their efforts into developing 2 distinct architectures (because they were already working in parallel at the 432 architecture for their future CPUs, which was a flop), so after developing the 8086 for the high end of the market they have just crippled it a little into the 8088 for the low end of the market.
Both 8086 and MC68000 were considered too expensive by IBM, but 8088 seemed a better choice than Z80 or MC6809, mainly by allowing more memory than 64 kB, which was already rather little in 1980.
In the following years, until 80486 Motorola succeeded to have a consistent lead in performance over Intel and they always introduced various innovations a few years before Intel, but they never succeeded to match again Intel in price and manufacturing reliability, because Intel had the advantage of producing an order of magnitude more CPUs, which helped solving all problems.
Eventually Intel matched and then exceeded the performance of the Motorola CPUs, despite the disadvantages of their architecture, due to having access to superior manufacturing, so Motorola had to restrict the use of their proprietary ISAs to the embedded markets, switching to IBM POWER for general-purpose computers.
Analysis of issues in making more performant 68k and VAX are major part of what led to RISC development, with complex addressing (even in earliest 68000) being part of the problem. People think of x86 as CISC when reading about CISC vs RISC, but x86 was not much of a consideration when industry was switching to RISC-style designs - it was hitting walls on complex ISAs, especially VAX (which was allowed to live for way too long), but also to an extent 68k.
N.b. 68000 was supposed to be a 16bit extension of 6800, which among others resulted in hilarious two layers of microcoding.
AS for IBM PC, 68000 had major flaw of being newer while 8086 had been available for longer and with second sources - 68000 was released at the same time as reduced capability 8088, while equivalent reduced capability model for 68k arrived in 1982.
68k did not resemble VAX at all, it was considerably simpler. 68k and the other Motorola CPUs resembled a lot the earlier PDP-11, not VAX.
Both MC68020 in 1984 and 80386 in 1985 have added to their base architectures various features taken from VAX, e.g. scaled indexed addressing. MC68020 has added slightly more features from VAX, e.g. bit-field operations, while 80386 has added only single bit operations. However none of the few features taken from VAX has made 68k more difficult to implement or less suitable for high speed implementations.
The wrong feature added in MC68020, which had to eventually be removed later, which consisted in the memory indirect addressing modes, was not taken from VAX. VAX did not have such addressing modes, only some much earlier computers had such addressing modes. Those addressing modes were added by someone from Motorola without being inspired by VAX in any way.
The VAX ISA was more difficult to decode at high speed, because it used byte encodings, like x86, but the VAX ISA was still much easier to decode at high speed than x86. The 68-k ISA, which used 16-bit encodings, was much easier to decode than x86, being intermediate in ease of decoding between a RISC ISA and VAX. The x86 ISA is probably the most difficult to decode ISA that has ever been used in a successful product, but at the huge amount of logical gates that can be used in a CPU nowadays that is no longer a problem.
The reduced capability variant of MC68000, i.e. MC68008, has been launched too late to be useful for IBM because Motorola had not realized that this is a good idea and they have done it only after the success of Intel 8088.
Simultaneously with MC68000, Motorola had launched MC6809, which Motorola believed to be sufficient for cheaper products. That was Motorola's mistake. MC6809 had a much more beautiful ISA than any other 8-bit CPU, but at the time when it was launched 8-bit CPUs were becoming obsolete for general-purpose computers, due to the launch of the 64 kilobit DRAM packages in 1980, which made economical the use of more than 64 kilobytes of memory in a PC, for which the 8-bit CPUs like Zilog Z80 and Motorola MC6809 were no longer suitable.
«More CISC-y» does not by itself mean «harder to optimise for». For compilers, what matters far more is how regular the ISA is: how uniform the register file is, how consistent the condition codes are, how predictable the addressing modes are, and how many nasty special cases the backend has to tiptoe around.
The m68k family was certainly CISC, but it was also notably regular and fairly orthogonal (the legacy of the PDP-11 ISA, which was a major influence on m68k). Motorola’s own programming model gives one 16 programmer-visible 32-bit registers, with data and address registers used systematically, and consistent condition-code behaviour across instructions.
Contrast that with old x86, which was full of irregularities and quirks that compilers hate: segmented addressing, fewer truly general registers (5 general purpose registers), multiple implicit operands, and addressing rules tied to specific registers and modes. Even modern GCC documentation still has to mention x86 cases where a specific register role reduces register-allocation freedom, which is exactly the sort of target quirk that makes optimisation more awkward.
So…
68k: complex, but tidy
x86: complex, and grubby
What worked for x86, though, was the sheer size of the x86 market, which resulted in better compiler support, more tuning effort, and vastly more commercial optimisation work than m68k. But that is not the same claim as «68k was harder to optimise because it was more CISC-y».
Notice I didn't write harder to optimize for - I am not talking about optimizing code, but optimizing the actual internal microarchitecture.
Turns out m68k orthogonality results in explosion of complexity of the physical implementation and is way harder to optimize, especially since compilers did use that. Whereas way more limited x86 was harder to write code generation for, but it meant there was simpler execution in silicon and less need to pander to slow path only instructions. And then on top of that you got the part where Intel's scale meant they could have two-three teams working on separate x86 cpu at the same time.
Once again – respectfully – this remains largely twaddle as the facts themselves state otherwise.
Even at the microarchitecture level, the hard part is not raw CISC-ness but irregularity and compatibility baggage. In that respect x86 was usually the uglier customer.
High-end x86 implementations ultimately scaled further because Motorola had less market pressure and fewer resources than Intel to keep throwing silicon at the problem, not because m68k was somehow harder to optimise.
Later high-performance m68k cores did what later x86 cores also did: translate the architected variable-length instruction stream into a more regular internal form. Motorola’s own MC68060 manual says the variable-length M68000 instruction stream is internally decoded into a fixed-length representation and then dispatched to dual pipelined RISC execution engines. That is not evidence of an ISA that was uniquely resistant to microarchitectural optimisation. It is evidence that Motorola used the same broad trick that became standard elsewhere: hide ISA ugliness behind a cleaner internal machine.
There is also a deeper point. The m68k ISA was rich, but it was comparatively regular and systematic at the architectural level. The m68k manuals show a clean register model and – notably – consistent condition-code behaviour across instruction forms. That kind of regularity is exactly what tends to help both compiler backends and hardware decode/control design. By contrast, x86’s biggest hardware pain historically came not from being «less CISC» than m68k, but from being more irregular and more burdened by backward compatibility.
Lastly, but not least importantly, CPU's were not the core business of Motorola – it was a large communications-and-semiconductors company, with the CPU's being just one product family within a much larger semiconductor business.
There was no clear understanding within the company of the rising importance of CPU's (and computing in general), hence the chronic underinvestment in the CPU product line – m68k did not see the light of highly advanced, performant designs purely because of that.
Well, here I am following what people who worked on CPUs at the time wrote.
And from the point of microcoded system like x86 and 680xx were (including 68060) it is important to how many microinstructions your instruction stream will decode - something that greatly favours ISAs that are not orthogonal - and major reason why x86 often has 1.2-1.6 ratio of microinstruction to instruction for overall program code.
Orthogonality makes it problematic because while it's easy in "interpreter microprogram" style of old and easy to program in assembly for, it means that for example for 68k you have to deal with many addressing modes for every operand - whereas x86 pretty much fuses it between 1 to 2 instructions because only one operand can have any computed address, and scope of available computation is limited (even compared to just 68000).
This means that while both architectures can use "translation microcode" approach, one (x86) will easily decode into one or two 72bit instructions (using P6 here) with worst case involving somewhat rare 3 operand form of memory address (which still can only happen for one operand of the instruction, not both)
Harder to optimize at microarchitectural level because each individual instruction represents way more complex execution model, including to even decode what the CPU is supposed to do.
X86 is comparatively simple, with limited indirect addressing support to the point it can be inlined in execution pipeline, and many instructions either being actually "simple" to implement, or acceptable to do in slow path. M68k (and VAX even more) are comparatively harder to build modern superscalar chip for.
The 68k family had only one bad feature, which was introduced in MC68020, a set of memory indirect addressing modes.
Except for this feature, all instructions were as simple or simpler to implement than the x86 instructions.
MC68020, like also 80386, was a microprogrammed CPU with multi-cycle instructions, so the memory indirect addressing modes did not matter yet.
Those addressing modes became a problem later, in the CPUs with pipelined execution and hardwired control, because a single instruction with such addressing modes could generate multiple exceptions in the paged MMU and because any such instruction had to be decoded into multiple micro-operations in all cases.
For embedded computers, backwards compatibility is not important, so Motorola could correct this mistake in the ColdFire CPUs, but for applications like the Apple PCs they could not remove the legacy addressing modes, because that would have broken the existing programs.
Besides the bad memory indirect addressing modes, 68k had the same addressing modes as 80386, except that they could be used in a much more orthogonal way, which made the implementation of a CPU simpler, not more complex.
For a corrected 68k ISA, e.g. ColdFire, it is far easier to make a superscalar implementation with out-of-order execution than for x86.
Like I have said, 68k does not resemble VAX at all. The base 68k architecture resembles a porting to 32-bit of the DEC PDP-11 architecture. Over the base architecture, MC68020 has added a few features taken from VAX, mainly scaled indexed addressing and bit-field operations, and a few features taken from IBM 370, e.g. compare-and-swap.
Intel 80385 has also taken scaled indexed addressing from VAX, but instead of implementing bit-field operations it has added only-single bit operations. That is a negligible simplification of the implementation, which has been chosen by Intel only because their instruction format did not have any bits left for specifying the length of a bit field.
None of these features taken from VAX has caused any problems in either the Intel or the Motorola CPUs in high-speed pipelined implementations.
> Those addressing modes became a problem later, in the CPUs with pipelined execution and hardwired control, because a single instruction with such addressing modes could generate multiple exceptions in the paged MMU and because any such instruction had to be decoded into multiple micro-operations in all cases.
This is literally my point - the people involved in shift to RISC had figured it was a problem, and one aspect that made x86 easier to optimize long term (outside of Intel's huge market share) was that x86 had at most one memory operand per instruction (with certain exceptions). m68k's orthogonality meant both decode and execution are long-term harder, especially since you're going to have to support software that already uses those features - x86 has less of a legacy baggage there by virtue of not being as nice early on.
Clean break towards simpler internal design backed by compiled code statistics led most vendors - including intel - towards RISC style. Intel just happened to have constantly growing market share of their legacy design and never committed fully to abandoning it while lucking out in their simplistic design making it easier to support it long term.
> but for applications like the Apple PCs they could not remove the legacy addressing modes, because that would have broken the existing programs.
The "FireBee" project which is/was an Atari ST-ish clone build on ColdFire ... proved this wasn't actually a huge obstacle, they made an Atari ST compatible system which provided a workable emulation mode that could still run classic 68k code. And then things like the kernel etc could be recompiled or modified in pure ColdFire mode for performance.
I have one downstairs ... tho, in fact, I never ever use it.
Certainly a lot less disruptive than the half decade or more of crashy slow 68k emulation in PowerPC that Apple was forced to do instead. That initial era of PowerPC Macs running MacOS versions prior to 8 were.. terrible.
ColdFire was/is literally that is my understanding. But there was really no market for it. They produced variants up to 300mhz if I recall but then relegated it mostly just to microcontroller market and then stopped developing it.
It was too late, and just oh-so-slightly incompatible with 680x0. But I suspect if the ISA used in ColdFire v4 had existed in 1994, 1995 that perhaps Apple's honestly disastrous foray through PowerPC could have been avoided.
Complicated, expensive CPU marketed to very high end workstation use? Nobody thought it was worth picking up even if it was faster than the alternatives.
Exept of course that Apple internally spent outrages amount of resources on their own CPU project that also wouldnt have had a volume story. Its only because that procet failed that they started looking at alternatives.
Apple really dropped the ball here. They had every ability to make something competitive with Nvidia for AI training as well as inference, by selling high end multi GPU Mac Pro workstations as well as servers, but for some reason chose not to. They had the infrastructure and custom SoCs and everything. What a waste.
It really could have been a bigger market for them than even the iPhone.
Just about everybody who isn't Nvidia dropped the ball, bigtime.
Intel should have shipped their GPUs with much more VRAM from day one. If they had done this, they'd have carved out a massive niche and much more market share, and it would have been trivially simple to do.
AMD should have improved their tools and software, etc.
Apple should have done as you say.
Google had nigh on a decade to boost TPU production, and they're still somehow behind the curve.
Such a lack of vision. And thus Nvidia is, now quite durably, the most valuable company in the world. Imagine telling that to a time traveler from 2018.
I think for AMD, they were focused on competing against Intel. Remember AMD was almost bankrupt about 15 years ago because of competing against Intel. But the very first GPU use for AI was actually with an ATI/AMD GPU, not an Nvidia one. Everyone thinks Nvidia kicked off the GPU AI craze when Ilya Sutskever cleaned up on AlexNet with an Nvidia GPU back in 2012, or when Andrew Ng and team at Stanford published their "Large Scale Deep Unsupervised Learning using Graphics Processors" in 2009, but in 2004, a couple of Korean researchers were the first to implement neural networks on a GPU, using ATI Radeons: https://www.sciencedirect.com/science/article/abs/pii/S00313...
And as of now I do believe AMD is in the second strongest position in the datacenter space after Nvidia, ahead of even Google.
> And thus Nvidia is, now quite durably, the most valuable company in the world.
Nvidia is the most valuable company in the world right up until the AI bubble pops. Which, while it's hard to nail down when, is going to happen. I wouldn't call their position durable at all.
The crashing and burning of Nvidia stock has been predicted for a while now and keeps not really happening. It’s gone pretty flat and volatile up there around $180 but they keep delivering the results to back it up. I was thinking this week that Apple is really primed to make a killing from people who want to run their LLM on-device coupled with an agent in the next couple of years. We’re a long way off being able to train the models – this is going to need an Nvidia-powered datacentre for the foreseeable future, but the local inference seems absolutely like a market that Apple could capture, gutting all the most premium revenue from Anthropic and OpenAI by selling Macs with a large amount of integrated memory to anyone who wants to give them the money to run their native OpenClaw/agent instead of paying ever-growing monthly bills for tokens.
It is definitely a case that they will fall a long way but Nvidia will not fail as a whole. They have a way of maximizing their position relentlessly. CUDA turns out to endlessly put them in amazing positions on things like image recognition, AR, Crypto and now AI.
For all the faults of them leaning in hard on these things for stock market and personal gains, Nvidia still has some of the best quality products around. That is their saving grace.
They will not be the world most valuable company once the bubble pops, will probably never get back there again, but they will continue to be a decent enough business. I just want them going back to talking about graphics more than AI again, that will be nice.
Why should Apple have done this? It doesn’t fit their business in anyway shape or form. Where does data centre hardware sit relative to electronics / humanities cross roads that is foundational for Apple?
Apple is presumably leaving a lot of money on the table by not trying to sell Apple Silicon for AI inference and training. They're the only ones who can attach reasonably large GPUs (M3 Ultra) to very large amounts of cheaper memory (512GB SO-DIMM per GPU). Apple could e.g. sell server SKUs of Mac Studios, heck they can sell M3 Ultra chips on PCIe cards. And they could further develop Apple Silicon in that direction. Presumably they would be seen as a very legit competitor to Nvidia that way, perhaps moreso than Intel and AMD. I'd assume that in the current climate this would be extremely lucrative.
Now, actually doing this would disrupt Apple's own supply chain as well as force it to spend significant internal resources and cultural change for this kind of product line. There's a good argument to be made it would disproportionally negatively affect its Mac business, so this would be a very risky move.
But given that AI hardware is likely much higher margin than the Mac business an argument could probably (sadly) be made that it'd be lucrative for them to try it. I personally don't think Apple is inclined to take this kind of risk to jeopardize the Mac, but I'm sure some people at Apple have considered this.
Yeah nothing about Apple is server side and imho that's what training is. To be serious about it as a company you have all sorts of other tools (crawlers, etc...) helping with training so it basically has to be in the datacenter at any reasonable scale anyway. And that's just not where Apple lives. We saw with Swift that they couldn't focus on server side enough to make it a serious language there and they've consistently declined to enter that area over the years because it's outside their wheelhouse.
Don’t mistake stock market performance for revenue. NVIDIA makes ~200B annually, same as what Apple makes from iPhones. It’s a big market but GPUs aren’t just AI.
I'm purely talking in terms of revenue. There's a huge demand for AI systems from personal workstations to datacenter servers, and Apple was one of the few companies in the world in a position to build complete systems for it.
But for some reason Apple thought the sound recording engineer or the video editor market was more important... like, WTF dude? Have some vision at least!
Their rule of only releasing major software updates once a year in June is holding them back IMO. Their local LLM apis were dated before macOS/iOS 26 was even released. Just because something worked 20 years ago doesn’t mean it works today, but I’m sure it’s hard to argue against a historically successful strategy internally.
One would expect the platform owner (especially one where they own both the hardware AND software) to provide a reasonable / easy path to using LLMs if they are going to provide a framework for doing so. But Apple can’t because of how slow they ship updates
It is more important. Both for the customer base that actually buys Apple machines as well as the cache and mindshare of being used by the people that create American culture.
Even if Apple had an amazing GPU for AI it wouldn’t matter hugely - local inference hasn’t taken off yet and cloud inference and training all uses servers where Apple has no market share and wasn’t going to get it since people had already built all the stacks around CUDA before Apple could even have awoken to that.
Apple already seems to do pretty well when it comes to AI systems on personal computers. Datacenters simply isn't their business, it would need some major changes on their part. Also, AI is a bubble, it will burst eventually, and because Apple doesn't have the fist mover advantage Nvidia has, they have a lot to lose entering this market now.
Sound recording engineers and video editors will not disappear after the AI bubble bursts, and Apple is wise to keep that market. Bursting the AI bubble will not make AI disappear, it will just end the crazy cashflows we are seeing now. And in that regard, with the capabilities of their hardware, Apple is in a pretty good spot I think.
Those back log orders are wild! One does wonder that if the bubble collapses or more global upsets happen in that time, how many of those will ever be fulfilled? Reality might be not so impressive, but considering if it fell even 80%, that is still $200 B in revenue and that is huge.
Remember when a $1 billion valuation used to be a big thing? That is nothing compared with nowadays.
They want to be able to sell handsets, desktops and laptops to their customer base.
Pursing a product line that would consume the finite amount of silicon manufacturing resources away from that user base would be corporate suicide.
Even nvidia has all but dropped support for its traditional gaming customer base to satisfy its new strategy.
At any rate, the local inference capabilities are only going to get cheaper and more accessible over the coming years, and Apple are probably better placed than anyone to make it happen.
Apple makes AI inference and training servers by the thousands. They just don't sell them to anyone. They use them internally in their datacenters. They didn't drop the ball, they are playing a different game while not cannibalizing their existing customer base.
Small models keep getting smarter and local hardware keeps getting better.
At some point, they will converge and an inflection for local LLMs will happen. Local LLMs will never be as smart or fast as cloud LLMs but they will be very useful for lower value tasks.
The Apple I was a pretty poor predictor of what mainstream mass-market computing was going to end up looking like. I don't think anybody has yet come up with the Apple II of local LLMs, let alone the VisiCalc or Windows 95.
If my Grandma had wheels she would be a bicycle. Apple would need to transition from being a consumer electronics company to being a B2B retailer for data centre hardware to take advantage of this.
Obviously Siri from WWDC 2yrs ago was a disaster for Apple. Other than that they seem to have done pretty well navigating the new LLM world. I do think they would benefit from having their own SOA LLM, but I don’t think its is necessary for them. My mental model for LLMs and Apple is that they are similar Garage Band - “Now everyone can play an instrument” becomes “now anyone can make an app”. Apple owns the interface to the user (i don’t see anyone making nicer to use consumer hardware) and can use what ever stack in the background to deliver the technical features they decide to.
> something competitive with Nvidia for AI training
Apple is counting on something else: model shrink. Every one is now looking at "how do we make these smaller".
At some point a beefy Mac Studio and the "right sized" model is going to be what people want. Apple dumped a 4 pack of them in the hands of a lot of tech influencers a few months back and they were fairly interesting (expensive tho).
> Apple is counting on something else: model shrink
The most powerful AI interactions I've had involved giving a model a task and then fucking off. At that point, I don't actually care if it takes 5 minutes or an hour. I've cued up a list of background tasks it can work on, and that I can circle back to when I have time. In that context, smaller isn't even the virtue at hand–user patience is. Having a machine that works on my bullshit questions and modelling projects at one tenth the speed of a datacentre could still work out to being a good deal even before considering the privacy and lock-in problems.
> What "tooling" do you use to let AIs work unattended for long periods?
Claude and Kagi Assistant. I tried tooling up a multi-model environment in Ollama and it was annoying. It's just searching the web, building models and then running a test suite against the model to refine it.
Cool? And it has nothing to do with what kind of consumer hardware Apple should sell. If your use cases are literally "bigger model better" then the you should always use cloud. No matter how much computing power Apple squeezes into their device it won't be a mighty data center.
For running the model once it’s been trained, all a datacenter does is give you lower latency. Once the devices have a large enough memory to host the model locally, then the need to pay datacenter bills is going to be questioned. I’d rather run OpenClaw on my device plugged into a local LLM rather than rely on OpenAI or Claude.
> At some point a beefy Mac Studio and the "right sized" model is going to be what people want.
It's pretty clear that this isn't going to happen any time soon, if ever. You can't shrink the models without destroying their coherence, and this is a consistently robust observation across the board.
I don’t think it’s about literally shrinking the models via quantization, but rather training smaller/more efficient models from scratch
Smaller models have gotten much more powerful the last 2 years. Qwen 3.5 is one example of this. The cost/compute requirements of running the same level intelligence is going down
I have said for a while that we need a sort of big-little-big model situation.
The inputs are parsed with a large LLM. This gets passed on to a smaller hyper specific model. That outputs to a large LLM to make it readable.
Essentially you can blend two model type. Probabilistic Input > Deterministic function > Probabilistic Output. Have multiple little determainistic models that are choose for specific tasks. Now all of this is VERY easy to say, and VERY difficult to do.
But if it could be done, it would basically shrink all the models needed. Don't need a huge input/output model if it is more of an interpreter.
There are no practically useful small models, including Qwen 3.5. Yes, the small models of today are a lot more interesting than the small models of 2 years ago, but they remain broadly incoherent beyond demos and tinkering.
I don't think you can make that case for 35b and up, including the 27B dense model. A hypothetical Mac Studio with 512 GB and an M5 Ultra would be able to run the full Qwen 3.5 397B model at a decent speed, which is more like 12 months behind the current SoTA.
A lot of people got a bad first impression about the 3.5 models for a few different reasons. Llama.cpp wasn't able to run them optimally, tool calling was broken, the sampling parameters weren't documented completely, and some poor-quality quants got released. Now that these have all been addressed, they are serious models capable of doing serious business on reasonably-accessible hardware.
Yes, but bigger models are still more capable. Models shrinking (iso-performance) just means that people will train and use more capable models with a longer context.
Give every iPhone family a in house Siri that will deal with canceling services and pursuing refunds.
Your customer screw up results in your site getting an agent drive DDOS on its CS department till you give in.
Siri: "Hey User, here's your daily update, I see you haven't been to the gym, would you like me to harass their customer service department till they let you out of their onerous contract?"
I’m running modest setup using a mistral model (24B) on a 9070 (AMD) and 32gb of ram. $1800 machine at the time I built it. It ultimately boils down to what you want to do with it. For me, it’s basically a drafting tool. I use it to break through writer’s block, iterate, or just throw out some ideas. Sometimes summarize but that can be hit or misss.
I don’t need the latest and greatest and I fine tuned LM studio enough that I get acceptable results in 30 to 90 seconds that help me keep moving ahead. I am not a software engineer, I am definitely not as much of a “coder” as the average person on HN. So if I can do it for less than $2000, I bet a lot of (smarter/experience coding) people could see great results for $5000.
You can get an M3 ultra Mac studio with 96gb ram for $4000. If you’re willing to go up to $6k it’s 256gb. Wayyyyy more firepower than my setup. I imagine plenty powerful for a lot of people.
Not all enterprises are the same, I imagine many companies have different departments working with local optimums, so someone who could benefit from it to get more productivity might not have access to it because the department that is doing hardware acquisition is being measured in isolation.
I think it’s a little unnecessary to lecture somebody on HN about how enterprises come in different shapes and sizes. It’s pretty clear what I’m implying here if you aren’t actively trying to assume the most reduced, least charitable version of my statement.
If Apple doesn't offer a Linux product, they cannot be used seriously in headless computing task. They are adamant in controlling the whole stack, so unless they remake some server version of macOS (and wait years for the community to accustom themselves with it), they will keep being a consumer/professional oriented company
How is this dropping the ball? I think they dropped the ball a long time ago by waiting until M5 to do integrated tensor cores instead of the separate ANE only which was present before.
For multi-gpu you can network multiple Macs at high speed now. Their biggest disadvantage to Nvidia right now is that no one wants to do kernel authoring in Metal. AMD learned that the hard way when they gave up on OpenCL and built HIP.
Apple didn’t drop the ball - they have no interest in creating servers for a limited time bubble. It is laughable that anyone would think that market would be bigger than iPhone - there’s a reason no RAM manufacturer is building new plants to take advantage of the current demand - they don’t expect it to last long enough to pay for their investment.
It's why WTI crude never exceeds $100/barrel. Every time it gets that high the insider in the administration shorts it and the administration announces a new policy delaying more strikes.
That’s how it worked 10 years ago, that’s not really how it works now. The wildcat shale drillers of a decade ago all went bankrupt when prices fell, and lost a bunch of money for their investors. They’ve since all been bought up by larger firms with way more capital discipline, who don’t ramp up drilling just because prices have a little spike, especially when we all know that TACO. Do not expect shale drilling to soften the blow of oil price increases this time.
You don't even need to trade oil, trading gulf country assets also works well. And rebuilding the gulf should be quite profitable if Iran blows it up and the assets will be available cheap.
Sure. If you go back 80 years and past several pointless wars you can find one example. Wherein the USA fought with a world wide coalition to defeat the Nazi's only after war had been declared by the Nazi's against the USA.
Actually, you can just look in Iraq, and CNOOC, PetroChina, and Zhongman. Companies will benefit, just not American ones.
And the US will clearly have failed to protect allies and project force in the Gulf. If the bulk of OPEC moves to a basket currency trade to ally more with PRC, India, and Russia that will be an astonishing failure.
I wouldn't have thought even Trump and this Republican administration was incompetent enough to break the petrodollar, but here we are, just one year in.
Also, don't look at fracking production curves. Bakken and Eagle Ford are foreshadowing the Permian.
I can see this war being mismanaged enough that gulf countries go to sell in yuans straight. After all, aligning with china will prevent further attacks from iran and almost all the stuff being made in the world already sells in yuans.
reply