
In the future
I can see Intel becoming a fab for customers, and developing some of their non-CPU tech like wireless chips etc.
x86 is dead, long live x86.
These are uncomfortable times for Intel and its investors. All the cool kids are talking about AI chips, an area where Intel has no real story to tell. Low power, embedded, and mobile have long been annexed by Arm. Now it looks like Arm is coming for Intel's most iconic territory – the PC. Last week, Nvidia, AMD, and Qualcomm …
is having to pay TSMC to make their GPUs, isn't a great sign.
Nor their just released and homegrown generation of x86, being almost indistinguishable from their last.
Still, got to be worth a punt on their currently collapsed share price - they're on the ropes and if this doesn't help bump them out of their decades-long-complacency, nothing will.
I very much appreciate that ARM has collected less baggage over its history than x86. On the other hand people have put enormous effort into optimizing x86. It takes the same number of transistors, time and power to do a multiply on ARM as x86. The big performance winner for Apple is HBM and having appropriate silicon to take advantage of the extra bandwidth. If Intel had a high performance ARM chip to plug into an x86 socket the speed and power requirements would be very similar.
Intel now have to take huge steps to produce chips that take full advantage of HBM without a helpfully timed fuck-ups from competitors - like AMD had with Itanium and RDRAM.
X86's problem is that to make it fast you need to have long instruction decoder pipelines, correspondingly complicated caches, etc. So the number of transistors the lie between an instruction arriving in the CPU and the ALU that's actually going to execute it is quite large.
Arm doesn't need all this. which is where it wins out. In an age when a good x86 core was needing millions of transistors, Arm was needing only a few tens of thousands for the whole instruction set.
US mega-corp, deeply ingrained in US political and business consciousness, threatened by emerging technologies that lack US-ness?
The obvious thing is not to compete on technology (as they're too late), not to compete on commercials (as they're too big and expensive), but to capitalise on the strengths any big US corporation has - political connections, playing the regulatory system, and litigation over anything that comes to hand. Potentially that may involve attacking US competitors, but the point is about ARM technology, and "not invented here". The non-US card is a very powerful one to play in a country where politicians are intensely nationalist (and easily bought), and courts are partisan.
Bog down the likes of Nvidia in slow moving legal and regulatory treacle, and it won't then matter whether IBM ultimately win the legal case, they'll have got what they wanted by gumming everything up for competitors for years. Very much like the Boeing/Airbus dispute that rumbled on for nearly twenty years, and was started because Boeing had in the four years leading up to 2000 gone from 70% market share to less than 45. The problem then wasn't really about price or subsidies (that the legal/trade case claimed) but that the Airbus product was more modern and simply better, and Boeing had nothing new to put on the shelves at that time - having taken their eye off the ball, buying and being borged by McD, and failing to sustain a technology lead they'd had for decades - Boeing didn't launch a genuinely new aircraft for sixteen years from 1995 onwards. The parallel with IBM is clear - a dominant player who've been complacent, got distracted, and find themselves challenged.
The problem with mega-corps is that they prioritize fintech type business over actual making of stuff. Its easy to see how this happens, its not some deep, dark, conspiracy but merely the quickest route to making profits and so 'enhancing shareholder value'. The US is littered with the carcasses of business that were once titans in their field but got taken over (often by a hedge fund) and stripped for value, leaving debt laden husk that eventually just withers away. When the business is large enough -- or is essential like a utility -- then the taxpayer ends up subsidizing it one way or another (either directly or through bloated 'aerospace' -- military -- contracts).
UK readers will be familiar with this since the country was a bit of a pioneer in this type of business practice.
Whilst agreeing with your general observation, I'd ask if that was what IBM did? Surely it was simply rank complacency at senior leadership levels?
I suspect many of us have worked for businesses where "not invented here" means that senior leaders are dismissive and disinterested. And that for me is the biggest challenge for any businesses, that once they're large they can't/don't/won't take the big risks involved in fast growth and developing breakthrough technologies. They see an influx of professional managers who do in fact know important stuff like how to administrate a large corporation, how to do stuff like regulatory compliance, how to be efficient. But they're not entrepreneurs, all too often they're not engineers or scientists, and almost universally they have diaries filled with low value crud, and they don't have the edginess to manage by wandering about. So instead of sitting through a half day on the quarterly update of the business plan, instead go and randomly disrupt the better sales bods and asking them what's happening on the front lines of sales, what customers are telling them, what competitors are doing. Or ambush the mid level engineers or developers and ask what's spoiling their day, or what they think of the company's strategy.
In my last job, my desk was situated about 40 feet from the CEO's office (UK business of a global major, UK turnover around £11bn). The bloke would scuttle into his office sideways like a crab, almost as though he was in fear of somebody speaking to him. In two and half years I didn't see him once socialise or exchange pleasantries with the various underlings who worked around him. Everything he knew was told to him by his direct (board) reports, and fully half of his information flow was Powerpoint, and he perpetually looked terrified. Unsurprisingly the company achieved nothing during his tenure. Curious thing was, I have reports from people I'd trust with my life that as a middle-senior manager this bloke was absolutely brilliant - thoughtful, friendly, approachable, considerate, and with a good brain.
Maybe there's a special course directors have to go on, and as part of that they have any courage, social skills, business acumen, common sense and likeability excised by brain surgery.
There was a study once which suggested that empathy (measured as the ability to figure out others emotions from pictures) decreased with rank in corporations - and since they collected data twice a few years apart they were able to get results from some of the same people following a promotion. Those people's results got worse. Something in our monkey brains turns off as you gain status, apparently.
> empathy (measured as the ability to figure out others emotions from pictures) decreased with rank in corporations
I would say that is more of a disconnect from meeting people the higher up the ivory towers they get.
Basically big corporations are like the huge dinosaurs that (allegedly!) took s long time to realise that the tail had been bitten off, they become huge and unwieldy with the turning response of a supertanker and innovation just seems to bleed out of them the more detached from reality they become
Business have to operate in manners which if they were human we'd call sociopathic and as a result they reward sociopathic tendencies in management
It has a habit of getting out of control once the upper levels are staffed mostly by such people and in peak cases you get Enron, etc
The problem with the Alpha is that DEC pulled all sorts of chip shortcuts to make it run so fast. They relied on the actual switching speeds of the transistor technology, and relied less on independent clock signals to keep everything in sync.
What this meant was that Alpha ran like greased lightening when it launched, and was also boosted by being a 64 bit architecture. It put the fear of god into IBM and HP at the time, who had been leapfrogging each other to have the fastet systems up to Alpha's launch.
The downside of this is that they had huge difficulty maintaining this gap in performance. But some of the cache tricks were the things that Intel illegally copied and then had to license as part of the patent litigation to speed up the Pentium processor. I never worked out how Intel got StrongARM in this deal, unless they were forced to buy it to generate cash to keep DEC afloat.
The WindowsNT port to Alpha was apparently very good, as was the (unreleased) PowePC version (which I saw when I worked at IBM, it was running on the same desktop hardware as the OS/2 port for PowerPC). Alpha obviously lost out when Compaq bought DEC when they ran out of money. Compaq continued to sell Alpha systems, but put no effort into continuing the processor.
PReP (Poert Reference Platform) systems were not that obscure in the IBM lineup of systems, although it was replaced by the CHRP (Common Hardware Reference Platform) standard quite quickly.
The system I saw it on was a pre-release 720-40P, which was a desktop RS/6000 with a PowerPC 601 processor that came before the much more widely known 43P desktop system. I had AIX 4.1 running on my system, unsurprisingly. The NT and OS/2 systems I saw were demonstrated by the marketing technical support team on the floor above me.
This was the first AIX system that I saw that did mutimedia sound and video. At one point IBM was intending to have desktop PowerPC systems as part of their line-up, and strangely, you can still see remnants of the Ultimedia Services in current AIX software!
How far behind the curve is the RISC-V - which already runs Linux in some incarnations? At the moment, a casual glance suggests that it's a lot cheaper than either x86 or ARM but I don't yet how well it integrates with common desktop/server components. But I know nothing about it in detail; I'm always eager to be educated...
A good guide to RISC-V's likely trajectory is to look at what's happened with MIPS. They're similar architectures. MIPS itself is actually a very good architecture, its lean and very fast but its suffered from it being somewhat redundant on the desktop so its niche was (is?) in embedded systems such as high performance network devices. This architecture is now relevant for all applications so I expect RISC-V to become a very significant player, potentially eclipsing even ARM.
Actually there have been Arm chips that can run Windows since 2012, maybe before then. These have been in-house chips which I don't think were publicly released. This whole NVIDIA Windows Arm PC news is based on a Reuters press release that everyone is parroting while adding a bit of extra flair to it. Nobody seems to know if this is actually happening or not.
Intel's problem is its own history. In the 1990's Intel got out of embedded and summarily pulled their embedded processors almost overnight, while they "encouraged" customers to switch to another supplier by marking up $7 processors to $25. Companies that could not switch fast enough went out of business because of the financials. Intel then spent $35 Million promoting the humiliation that was XScale which was rejected by an untrusting industry still smarting from their embedded exit. Since then Intel has quickly introduced a range of embedded products which they then quickly pulled from the market (Atom, Edison, Galileo, et al) each with subtle technical flaws and very poor support from Intel. This distrust of Intel for embedded has not died out even 30 years later - why source a product that could be discontinued next year?
Meanwhile RISC-V is snowballing. Talk to an insider at any major semiconductor company and they whisper to you they are in development of a RISC-V product line, including NVIDIA. RISC-V development boards are very popular. At least two semiconductor companies that I know of are working on Windows RISC-V processors and I'd put a bet down in Vegas that a prototype RISC-V Windows PC will be available within the next few years while people are still talking about Arm PCs with "-ing" words (working, developing, planning, etc.).
The ugly industry truth is Arm has no strategy to combat RISC-V because they have no significant internal strategy to combat it other than a tepid PR strategy and the NVIDIA Reuters article may be part of that tepid strategy. Its no secret Softbank wants its money back out of Arm, right now, and to that end Arm's R&D is suffering. Arm appears to be trying to slow RISC-V development using their proxies by attempting to fragment the architecture in RISC-V committees but they are failing. RISC-V ̶m̶a̶y̶ ̶b̶e̶ will be dominant in 3 to 5 years, and will eventually replace Arm.
The only sense in which RISCV is snowballing, is going downhill and taking a lot of bodies with it. SiFive just threw in the towel. Laid off its RISCV staff, stopped all its own RISCV chips and cores. Their “pivot” is that they still have some engineers, and if *some other fool* would like to pay them to customise a RISCV core, they are happy to take the money.
Nobody will, of course. There is no business to be had here. They’ve tried “sell a RISCV IP core”, and that doesn’t work for reasons I’ve explained in the past. They’ve tried “make and sell a RISCV chip” (Horse Creek) that was Dead On Arrival as a business plan, even before Intel tried to “help”. And now they’re trying “sell our expertise to other companies”…..which doesn’t work, because it’s open source, so there’s no IP, and anybody can hire the same engineers and cut out the middleman.
It’s the same as it always was. Anyone can build one. But unless you enjoy spending several hundred million dollars on your vanity hobby, there’s simply no way to generate a profit….which means there’s no way to build chips in volume for customers. Soon enough, VCs figure that out, and stop throwing money. Which they now have.
> SiFive just threw in the towel.
From The Register, 5 Oct 2023:
"Last year we learned that SiFive's RISC-V-compatible CPU cores would power NASA's High-Performance Spaceflight Computer (HPSC).
"The computer system is being developed in collaboration with SiFive and Microchip under a three year $50 million contract. When complete, the part is expected to become the backbone for future manned and unmanned missions by space agencies."
https://www.theregister.com/2023/10/05/riscv_microcontroller_space/
That is a wildly optimistic estimate of the progress Riscv will make. Arm is almost everywhere already, Windows and MacOS already runs on available machines. Microsoft even showed of windows 7 and office running on an Arm system back in about 2008; 15 years ago.
Windows won't run on Riscv unless microsoft make it do so.
Arm's best strategy is to not give customers a strong reason to move off the platform. The price they charge is pretty low anyway, which is partly why Arm is so popular. Chip houses are certainly going to take a good look at Riscv, but it's far from certain that it's worth abandoning Arm for minimal gain and a whole lot of market risk.
Arm don't need to foment fragmentation of the Riscv ecosystem, it seems pretty good at doing that itself.
ARM is really just "half a RISC-V". It merely has much more momentum because there's a lot more to a processor than the CPU part and its undergone extensive development over a period of decades. (ARM knows about this only too well -- their early standalone processors were not just slow, they were dire.)
What's going to push RISC-V development along is the US government. Thanks to entity lists, sanctions and general interference with normal commerce the Chinese know that using ARM has no future -- technology you can't control can always be used against you. This doesn't mean that you'll see top of the line, world beating, RISC-V parts next week, what's likely to happen in the short term is that all those commodity processors, the SoC parts that have been the bedrock of ARMs business, will be replaced. Then, as with ARM, the parts will work their way up the performance table.
As for whether it can run Windows or not, no 'adversarial' country (which we've deemed China to be) will want to run an American OS. You need control over what your system is doing, you don't want it to be phoning home all the time. Anyway, Windows is getting more and more erratic and by extension less and less useful. It really has passed its sell-by date. (Anyway, there's always emulation if you're desperate and that at least gives you a controlled sandbox to run the code in so you can keep an eye on what its doing.)
That is a highly revisionist view of how ARM got to where they are today. ARM's earliest CPUs were lightning fast in comparison to PCs of the day. I know, because I used them both.
Since the days of the Acorn Archimedes, ARM has always designed CPU cores to match the market segment interested in using it which was an awful lot microcontrollers and small application processors. There's is not to design final products themselves; that's not their business model.
It's only comparatively recently has there been a market for bigger CPUs with higher performance, first in the mobile devices sector and now increasingly in desktop / laptop and server. And, this is not necessarily ARM developing these; they've licensed the ISA to companies like Qualcomm, Apple, Samsung, Amazon, Google, etc. and it is they who are designing the large scale devices. ARM isn't, and doesn't have to; they just sit at the top acting as a paid referee ensuring that when companies put "ARM" on the lids of their chips they will indeed run ARM op codes as expected. Risc-V cannot easily compete on price - the difference between "very cheap" (Arm's license price per part) and free is not going to improve the bottom line much.
If Arm continue in their role as "paid referee" to most manufacturer's satisfaction, they're probably going to be around for a long time. The recent IPO in the US appears to have been successful, and that's a big vote of confidence (that's a lot of US investors giving money to Softbank for a company many said they'd overpaid for; a lot of investors really, really like ARM).
It's interesting to see the Risc-V foundation trying to mould its ecosystem towards the kind of disciplined consistency that Arm has (the difference being that Arm has the contractural sticks to be able to do that, the Risc-V foundation can only ask politely).
China cannot drive Risc-V to the same levels of performance on purely architectural design prowess. They'd need access to the same silicon processing capability that TSMC has. The US is being firm on denying them that access.
Whilst there's business-critical software running only on Windows, China and everyone else needs Windows. You may or may not consider Office to be business critical, but strategically important software such as Catia seems to be solidly Windows-based. And on the reliability front - I have more trouble with running Linux bare metal than I have running Windows. I prefer to run WSL rather than bare metal Linux.
"a lot of investors really, really like ARM"
This is incredibly dangerous for am IP company like ARM and frequently results in investor demands for "profit" pushing licensing costs up
In addition the USA's ban on ARM selling to chinese companies is rapidly driving RISC-V development (which in turn will caise companies to switch away from ARM and cause spooked investors to demand licensing fee hikes as volumes go down, which accentuates the problem)
It's a variation on the theme of "The Internet interprets censorship as damage and routes around it"
"Arm is almost everywhere already"
The problem ARM has is that with VCs/Bankers at the helm, they're trying to maximise income by tightening licensing conditions and increasing prices, based on the idea that "everyone uses ARM, so we're dominant"
This is a fast way of driving customers into the loving arms of compeitors (no pun intended), as we're seeing.
The only consolation is that if Nvidia had gotten hold of them this would have happened even sooner and RISC-V wouldn't have been ready for prime time
Even MIPS Inc has a RISC-V core now
Every company is trying to maximise income. That's their job.
Don't expect any company making RISC-V based cores to behave any differently, or to somehow have not-VCs and not-Bankers at their helm. The RISC-V ISA may well be open source, but that's essentially pointless. You can't as an end user make your own chips or hardware, you'll have to grease the palm of someone who needs to make a profit to get hold of them, and they'll be looking to maximise that profit no matter what. If they think they can make more money by adding in exclusive bespoke extensions in the hope you get addicted to them, they will.
ARM may or may not now have a management that understands ARM's position in the market. If they do try to gouge the market they'll soon learn the error of their ways in the shape of anti-trust actions. That's what happens to dominant abusive CPU ISA owners, especially in the USA. The precedent is Intel, who were obliged to license x86 to competitors (including AMD) at a reasonable rate. Indeed, the reason NVIDIA did not buy ARM was because the competition authorities all over said "no way" (plus a bunch of other objections).
Also, Softbank Group is not a venture capital fund, and it is also not a bank. Softbank Group is an investment holding company. It retains 90.6% of ARM having bought back just before the IPO the 25% of ARM that it had previously sold to the Softback Vision Fund (which is a VC fund). As an IHC it's intended to be in it for the long term, not the short term that a VC is interested in. Softbank Group is perhaps not the best or wisest IHC, but the IPO and the transfer of 25% from VC to IHC could be taken as indications of an intent to make money eventually, not quickly. The "eventually" is be welcomed, indicating an intent to preserve ARM's role more or less as we currently know it. If they screw it up, so be it; but they don't have to try very hard to succeed (just by not upsetting the apple cart...)
There is a very big difference between long-term and short-term value.
Most of the ways to maximise the total profit over the long term involve investment, thus lower short-term profit. They're also difficult, requiring good knowledge of the markets, and carry short-term risk.
Most of the ways to maximise short-term profits are highly destructive and will leave the company as an empty, negative value husk in a few years. But they're easy to do and require little, sometimes no market knowledge.
When the owners expect to have sold up and left within a year or two, which option will they choose?
VCs/Bankers are not at the helm (yet) at ARM. Only 5% of the stock is traded, with Softbank still being in total control. The stock market listing has provided a 'value' for ARM so the real fun starts as and when Softbank starts selling large blocks of shares to interested parties. Then we see who will be running ARM and what direction it takes.
MIPS mostly suffered from restrictive licensing conditions and this is why when Chinese manufacturers had an opportunity for a royalty-free CPU in RISC-V, they grabbed it
(Early chinese MIPS-compatible implementations avoided 4 load-store operations for this reason and it was STM who paid for Longsoon's license from MIPS inc)
Bear in mind that MIPS and RISC-V were both open academic projects. The difference is that MIPS Inc was spun out to commercialise that architecture and did it in ways that Americans know best (locking things down)
It's not entirely fair to say that Intel has a deep history of not innovating.
Sure, they've repeatedly screwed up every new ISA they've tried since 32-bit x86, needing AMD to get them successfully into the 64bit era. However, they were kings at silicon process innovation. Who needs a different ISA, when you can blitz everything else using the old ISA but on an improved silicon processing node? Intel successfully kept that up for decades. This proved insufficient in the mobile arena, but it continued to stand up to scrutiny in the desktop / server market. And it still does - except that it's AMD and they're leveraging TSMC's superior silicon process prowess.
The problem was that they got into the US's stereotypical hire/fire ways with their silicon process designers / engineers, fired them all and then wondered why they couldn't make 10nm work for them at all whilst TSMC waltzed off into the 7, 5 and now 3nm distance. Until they can get back into silicon process dominance like they used to have, the ISAs / chips they have to offer pretty much doesn't matter at all. For instance, an ARM built on an Intel second-rate silicon process is going to run slower / hotter than an AMD-designed ARM SOC build on TSMC's superior process
Regarding the ISA itself, there's precious little point innovating. ARM has left very little room between their designs and any notional "ideal". MIPS and RISCV aren't significantly better from an ISA performance point of view. If Intel created yet another ISA, it'd be doing it or no good reason.If they want to get on some kind of par, licensing ARM and just aiming to match TSMC would be a good start.
there's precious little point innovating
The main innovation of RISC-V is its licensing and I can see some merit in that. But an instruction set isn't a chip design and the people with experience of producing the silicon start with a huge advantage - as well as a lot of IP in other areas of the process and customers' sunk costs in tooling and training. There are lots of ways for established suppliers to snuff out the upstarts including, if necessary, producing their own RISC-V chips in quantity.
I'm unconvinced about that. Sun used to publish the SPARC CPU designs, and countries like Iran made their own versions. There's numerous GPL-licensed SPARC designs. Some of these date back to at least 2006, rather earlier than RISC-V. RISC-V is just another one, arguably with better publicity, and arguably more closely aligned to mobile applications than other open source ISAs and their designs.
And this should be part of the lesson. We've been there before. It made no noticeable difference. Dominant ISAs dominate regardless of the availability of other ISAs, until there is a good enough reason to adopt another. People didn't suddenly rush off and make SPARCs or POWERs to break the dominance of x86/64; it wasn't worth it. ARM succeeded only because it was a much better fit for mobile devices than x86 at a time when ISA power efficiency really, really mattered to get any useful functionality at all. Now we can put 15 billion transistors in our pockets (that's how many an Apple A15 has), easily enough to implement a pretty potent version of any ISA we've ever had (even x86), though ARM have been (to date) careful not to give anyone a strong reason to do so.
Several of the key players in the mobile industry making their own ARM devices are also members of the OpenPOWER foundation, and nominally have unfettered free access to that ISA, even the very tiny Microwatt version, but haven't bothered to shift from ARM to it. Some of these companies are also founding members of the RISC-V foundation. If Apple, Samsung, Google, NVidia or Amazon wanted to save a few bucks by losing ARM's license fee, they could have long ago but haven't. Even though they are perfectly capable of taking ISAs like POWER (which they have access to) and re-engineering them for mobile applications. I note that Google have begun a RISC-V port of Android, but they're not exactly rushing to make a chip themselves for it to run on...
There are at least two good reasons for ditching a big hairy old ISA:
1) It's far quicker, easier and cheaper to design an implementation of a clean, small and well defined IDA vs a very complex very old & crufty ISA.
2) It's far quicker, easier and cheaper to validate an implementation of a clean, small and well defined IDA vs a very complex very old & crufty ISA.
Those two reasons underpin why RISC architectures continue to survive and thrive through domination of the SoC scene - which happens to be where most of the money is. Shipping big clunky and expensive 2000+ pin packages is dandy - but it doesn't cut it when folks are trying to sell a couple of million mobile phones.
The most damning indictment of Intel's innovation failure is that it was *AMD* who developed the current dominant incarnation of x86 (AMD64 - remember that ?). Just to add a bit of salt to the wound there were senior Intel engineers posting on USENET sometime before Itanic (2001) saw the light of day that stated they reckoned a 64bit cut of x86 was what folks wanted (and was very doable). Not to mention that the whole dynamic / static optimization argument had already been decided by the Alpha EV6 (1998) vs the EV4 (1992).
And the one good reason for not ditching a big hairy old ISA is there's no point doing so if the software devs aren't going to follow you (as that senior Intel engineer seemed to know!).
For me the most damning thing about Intel is that, having the world lead in silicon processing (as they used to) they could have cleaned up with an ARM license. Imagine what Intel could have done with an ARM design, back when Intel's transistors were the best, smallest and cheapest in the world. If Intel had decided to dominate the mobile space by leveraging ARM's designs and their own silicon fab prowess, they'd have cleaned up. Innovation level? Zero (well, limited to their silicon fab tech). Profit level? Sky high, and it's profits that count, not "innovation".
They might not have enjoyed writing "ARM" on top of each and every chip produced, but as they'd have owned the desktop, server (x86) and mobile (ARM) markets I think they could have come to terms with that slight dent in their pride. But no, they thought they knew best, they thought they could swing the mobile world back to x86, etc.
Intel forgot that it is, fundamentally, a purveyor of transistors. It makes money selling transistors. If it's wiring them up in ways customers don't want (e.g. as x86's), and also isn't making the sort of transistors that people want to buy (e.g. they're hot and slow), then they're not going to sell very well.
BTW, 2000+ pin packages; the A15 from Apple isn't too far behind that, about 1000+, looking at some pics on YouTube. For a SOC that is supposed to have more of the devices electronics all on one chip, it's got an awful lot of external contacts.
Intel did have an architecture licence, and look what they did with it; XScale - the worse ARM core ever. They took all good work DEC did on the StrongARM and threw it away, coming up with a chip with a hideously slow memory system, a core they clearly did not understand, littered with unnecessary barriers crippling the pipeline, and a huge number of bugs - the errata was larger than the entire manuals for any of ARMs own designs.
It wasn't that Intel sold off the ARM division because it didn't know what to do with it, but I think it was more the embarrassment of doing such a bad job.
Oh Nvidia is making CPU Oh X86 in trouble, Nvidia is making CPU on ARM for years, it is inside the switch, so where is ARM desktop?
Oh Qualcomm is making a powerful CPU on ARM X86 in trouble, how many years did Qualcomm made desktop CPU, where they are now, nowhere.
Oh XXX is making a very powerful ARM server X86 in trouble, how many years was ARM in server.
Oh Facebook is now using ARM X86 in trouble, what is the % of ARM in Server.
Oh Apple is dropping X86 for AXX chip X86 in trouble, 3 quarters later, mac share of desktop is "Decreasing"
ARM is not power efficient X86 is just as efficient as ARM, RISC Vs CISC is long dead. The transcoding now days is less than 0.5% of die area i.e. you can't felt any difference, plus inherently X86 have much better compilers GCC, LLVM is running X86 as good as ARM, can't see why ARM can move to desktop.
Windows NT4 and 2000 ran on ARM (and MIPS, SPARC, and PPC) then they dropped support for everything except x86 with XP through 7, unless you count the Itanic versions, but then Windows 8 brought ARM support back in the form of Windows RT. So, internally they probably never really dropped support for ARM, maybe they were even (very) quietly selling it for embedded platforms. Windows 10/11 for ARM is out there as well, but currently only in OEM form.
I would agree that Windows on ARM is " for developers and enthusiasts", however you can buy at least one ARM laptop from a Tier 1 supplier with Windows installed (Thinkpad X1s).
From the Lenovo web site "13.3" super-responsive Windows with Snapdragon® powered laptop & ISV-certified apps"
This site reviewed the X13s - https://www.theregister.com/2023/03/21/lenovo_thinkpad_x13s_the_stealth/
When discussing Intel's notable failures, it would be remiss to omit the iAPX432. It was the 32-bit processor that was supposed to replace the 16-bit 8086 family of processors.
It flopped so hard it was detected by seismographs the world over.
It was ultra-CISC with all sorts of high-level constructs directly implemented in an instruction set designed explicity for running code from high-level languages like Pascal and Ada.
However, it turned out to be way slower at running real programs than the 80286 that was already in use widely when the iAPX432 was released.
Critics love to point out Intel's failures, but the fact that Intel is one of the most successful technology companies in the world indicates that it has overall made successful choices. Every major company has a raft of failures under its belt; the question is how Intel will meet the challenge by Arm on Intel's home turf. Ironically, the AMD story may point the way: Intel may manage to successfully ape whatever Arm does and thus keep Arm in eclipse. AMD, although it's been a thorn in the sides of both Intel and nVidia, has never actually managed to acquire more than a bare fraction of their revenue or market share.
Either way, we as consumers should ultimately benefit from increased competition and the innovation which results, so I'm looking forward to that outcome.
I have the impression that x86-64 is in an awkward position between RISC and CISC where it doesn't optimize well. An option for Intel might be letting ARM win the RISC market and go for a new CISC design. Create a high level instruction set that can do common computation, AI, and graphics work. Intel can leverage their expertise on ultra-complex processors to put JIT compilers into a broad range of chips. They'd all have varying hardware abilities but support one instruction set. It might actually optimize very well.
x86 seems like a dead-end.
At their core, all x86 CPUs for quite some time have been RISC. They translate the CISC instructions to RISC and then process it. But, yes, x86 seems like it's on its way out. As soon as someone releases a GP ARM CPU that has PCIe interconnects for external video cards, that'll be pretty much all she wrote for x86. Given nVidia's graphics card business, I'd bet money on that being part of their design.
Itanic was designed to optimize well. It didn't succeed.The compilers were poor from the get go and never improved much despite (in theory) being able to do much better. It was like watching a sports game where there's a high ball and everyone ran away shouting YOURS ....
So I don't think that's a valid argument, I've programmed in C and asm on x86, Power, Z , Sparc and ARM, even PA-Risc. Really there ain't much in it. Yes, some run some sequences better than others and there's the occasionally 'oops', but those don't make much difference overall between supposedly RISC and supposedly CISC machines.
The difference is more how many variants of CPU do you want you code to run on and the tradeoffs there. Apple have it good, they own their ARM arch, they own the applications and OS so they can turn the knobs up to 11.
You can't do that on x86 unless you recompile each time (O.K. there are some tricks in libc on Linux where paths switch according to underlying processor) but it doesn't help enough. Even in the Open source world almost no one goes to the effort of recompiling everything to run native optimized.
They've already tried that several times: iAPX-432 (early 1980s); i860 (1989-early 90s) and Itanic (early 2000s). Great ideas in every one of them, and all fatally flawed.
Now you mention it, perhaps they're overdue for another attempt — although the last one took an awfully long time to die. RISC-V might be their best bet, if they don't want to give in to ARM.
That seems to sum up Intel for the last few decades. Granted with the Atom it did eventually lead to the Centrino line of chips, which for the time, were pretty power efficient, but it took them quite a while to get there. As pointed out in the article, Intel expects to be able to just bully the market into going whichever direction they choose, and maybe once upon a time that was the case, but not anymore.
If I were Intel, since I don't have an ARM license or anything anymore, I'd probably be betting big on RISC-V. The base ISA is open so they could try to step in and be a real leader in the direction it takes. As well as being a large vendor for RISC-V chips that could potentially counter ARM. They could also fill capacity at their fabs by making RISC-V cores for other companies. The only other option would be to go crawling back to ARM and try to negotiate a new license. Doable, but I just don't see it being viewed as a viable option for an American company. It would mean admitting the top company executives were wrong, and that's like the one thing American companies will never do. Doesn't matter if the CEO is the last one shutting off the lights and locking the doors because the company went titsup, they'll still try and blame external market forces of some form or another for the failure, not that they made a bad decision.
I think the likelihood that RISC-V will take hold in the PC space is greater than ARM. The RISC-V ecosystem is growing by leaps and bounds and hordes of Chinese chip manufacturers are storming the barricades to evade trade sanctions imposed on ARM. How long before they reveal a "home-grown" PC with a RISC-V CPU?
ARM is already hurting from RISC-V on the low-end and is therefore attempting to move up the value chain. Too bad they'll find RISC-V waiting for them there too.
ARM probably doesn't have too much to worry about unless you start seeing RISC-V based phones and tablets. Maybe streaming devices too. If Google decided tomorrow that RISC-V was the future of Android, I'd be worried if I were ARM. Same as if Microsoft announced that they'd be releasing a RISC-V port of Windows and it'd be a full-fledged citizen alongside x86 and ARM. I seem to recall the RPi devs saying how they looked into RISC-V for the RPi5, but couldn't really find anyone licensing suitable cores. So, while it never pays to be complacent if you're the top vendor in a market, I think right now ARM is still in a pretty good position. Who knows what tomorrow will bring. Maybe RISC-V will sputter and fail, maybe it will overtake ARM, maybe it'll never do much more than nibble around the edges. Still, it seems like it'd be Intel's best bet if their pride prevents them from obtaining another license from ARM.
That a RISC-V smartphone will appear in the near future is almost a certainty. And Google doesn't decide if RISC-V is the future of Android, the vendors and the consumers make that choice for them.
If all Chinese smartphone makers start building RISC-V based devices then you're talking about hundreds of millions of SoC's using the ISA. That's enough for even Google to switch gears. Also, low-cost MCU's will undercut their ARM rivals by a substantial margin.
This article has all the fury of someone being salty about seeing 25% of their initial investment being written down by the market. Arm is getting eaten alive by Risc-V in the micro controller space and is now forced to be more flexible on licensing in SoCs, so having their iP in a few more locked down laptops (if they sell, surface on Arm never did so well) won't juice their quarterly results either. So after a few disappointing earnings their P/E will likely come right down... In the mean time Intel will likely make some nice margins on fabing those Qualcomm and NVIDIA Chips. So could be worse... Like most british tech things, they'll wither away when exposed to the harsh winds of the market. I just hope that Masayoshi Son gets most of his money back before it fizzles.