704 comments

[ 3.1 ms ] story [ 334 ms ] thread
I know it says “more” than $32B but isn’t that a little low if Softbank paid $31B?

Neither Softbank nor Nvidia care what I think, but I would feel better if the buyer of ARM wasn’t a company with a existing chip business.

Why do you think ARM is undervalued? While I agree, that considering the value of the ARM Arch brought to the mobile world, it may seem undervalued when you compare it to the insane valuation of some BS software companies, their business model of licensing IP doesn't allow them to make mega money. Apple and others have a perpetual licenses even.

Sure, if they want to, they could change the licensing model to bleed their customers dry and increase share price but that would just work temporary as everyone would then accelerate the movement to RISC-V.

ARM(the company, not the Arch) has peaked and will probably just stagnate for the foreseeable future.

I just find it hard to believe that Softbank would settle for $1B in profit for a highly successful business. They may have overpaid themself or ARM is losing them money, but why would Nvidia want to pay $32B for a money losing business?
Because Softbank itself is hemorrhaging money/losses and needs to start protecting itself.
Exactly. They have lost so much money - this will at least put some back in their coffers - how much did they lose on WeWork? $18B
They probably have no idea how to hype up ARM. It's like I bought out Apple Store in NY in cash but have no idea what to do with it other than to admire it
The multiples arm is trading at and With the new Apple lines I think they are doing a good job out of it .
SoftBank is run by an idiot that he paid 10x revenues for ARM when its growth was already slowing substantially.
If you look at the products these chips are in and see the breakdown. Cost of IP license from ARM, price to make that chip and price that chip sold in that product, let alone the final product price. The margins are thinest in ARM's area and for each chip they sell, the others down the line do have larger margins. I don't have TMSC's production costs and what they charge at hand, but i'd hazard that what TMSC make per chip is possibly more than what ARM makes per chip.

Maybe why SOftbank want to raise those licence costs and that is with alternatives now about, not that easy an equation than just raising those costs as many for controllers that RISCV become more than fine for and some HD manufacturers already transitioning for a few cents extra savings currently, let alone if ARM license increased.

Sure ARM here there and everywhere but it got that way as much on the cheap costs of using that IP as much as the array of IP packaged up. So yip they make money, but it's if you breakdown how they make money, ARM are regular/reliable, hence less spikes in income either way.

Logicaly, for ARM to increase revenue, it would need to branch out into other markets. Nvidia honestly would be a good fit for that over shafting licensing costs, which may well see an increase, but for what softbank wants for returns - ARM would not wear out the level of increases they would want and Softbank know this - hence sell or IPO them best for Softbank and also ARM.

> I don't have TMSC's production costs and what they charge at hand, but i'd hazard that what TMSC make per chip is possibly more than what ARM makes per chip.

Most ARM chips made aren’t pushing 7nm ultra expensive processes - and you are probably still correct.

That’s good take on ARM’s margin per chip.

(comment deleted)
If ARM could raise licensing fees, it already would have.
Also keep in mind that Softbank has needed money to pay off their large debt load, which probably reduces the amount of leverage they have in negotiations.
No, if anything it's overvalued. Previous year's profit was just over $400m. At a 10x multiple that's a $4bn price tag. If you go the year before it was over $600m, which at a 10x multiple is $6bn. Even a crazy 30x multiple would only be $18bn. Softbank massively overpaid, which is why Softbank is SELLING Arm and other companies, because they racked up massive debt.
Revenue is the best indicator for value, not profit which is artificially driven down for tax purposes.
Free cash flow is better than revenue - you can play around with accruals in unsavory ways that are harder to pull off with cash.
Neither is a great indicator for companies that are experiencing significant growth. ARM Holdings has nearly tripled revenue over the past decade. So you have to weight future growth, as well as potential future market opportunities.

ARM + nVidia can be a powerhouse combo, especially in the cloud/server market.

NVidia is already an ARM licensee. What does owning ARM holdings give them besides a massive amount of debt? They could license everything ARM owns for decades for less than this buyout would cost.
ARMs growth has petered out, its market is large and mostly saturated. PCs and servers won’t ship a fraction of the CPUs mobile does.
That's just silly, Wal-Mart would be the most valuable company (while being far from the most profitable) if it held true.
I don't think it's that silly to consider Wal-Mart the most valuable company.
But its definitely not in any financial metrics. If you own walmart for a year, you'll make around 4 billion, while if you own Google, you'll make around 7. And Google is likely going to be growing faster than Walmart.
If I sell 100 lemonades per day and can't pay my rent, yet my neighbor sells 50 lemonades but manages the business in a way to pay his rent and a salary, how can my business possibly be more valuable? It's a silly example but I don't think it's far fetched. Similarly, Apple is worth far more than 20% of the phone market despite having a grasp on only 20% (give or take, I don't recall the exact number) market shares. There's a reason companies aren't valued based on revenue.
Revenue is a terrible indicator, EBITDA is a good indicator.
Interest rates are at a record low. Amd is at a pe ratio of like 100, the average pe ratio today is around 15 iirc. With a growth expectation increasing because of apples adoption and all, plus with record low interest rates, a 60x multiple is not insane.
60x might not be impossible, but it's absolutely insane, even if it's common.
Nvidia's own pe ratio is around 80 it looks like. Its not actually that insane. Overvalued? Maybe. But all this is saying is that basically, you think that your current required rate of return minus your expected growth rate for ARM is around 1.7%, and your required rate of return is going to be pretty low right now because interest rates are so low. You value future earnings more and you think it's going to grow a lot. You might be wrong, and for a company like AMD with a pe ratio above a hundred you might be betting in an awful lot of growth happening, but in this environment, these are not "insane" numbers.
P/E ratio is not the same as a purchase-valuation multiplier.
What you brought up above in this thread is how expensive the acquisition looks as a multiple of earnings. That multiplier is the price of the company vs the earnings for the company, which is literally just a simplified P/E ratio that ignores dilution etc. You may be thinking of revenue multiple based valuation instead?
Apple has a one time ARM architecture license so that’s not going to be recurring revenue for ARM.
This is more from a increasing adoption of arm industry wide perspective as people begin writing code on and optimizing code for ARM machines. Could trickle over into increased adoption by other companies and other sectors.
They pay a royalty for every chip they ship.
> the average pe ratio today is around 15 iirc

I'm not sure what population of companies you're taking your average from, but the S&P 500 p/e is a lot higher than that.

That was a mistype, thanks for catching that. I meant 25, but I haven't kept up with the last quarter so I have no idea what it looks like with all the recent insanity, so that might be pretty far off as well
ARM the companies growth rate has been tepid at best since SoftBank bought it. You need a high growth rate to justify a high multiple.

And interest rates can change quickly. It’s a value trap to overweight current interest rates in equity valuations.

> ARM the companies growth rate has been tepid at best since SoftBank bought it.

It's a "knock it out the ballpark" victory by Softbank's standards.

If that happens it may have wider consequences.

ARM would become owned by an American company. It has to comply with American restrictions on dealing with China but that would make it essentially no different from any other American company, either legally or in fact.

So in the current context this would be bound to raise eyebrows in Beijing, and China could only react by doubling-down on developing domestic alternatives.

I'm kind of wishing this trend will push investment in something open like RISC V. Up and coming countries should definitely be thinking - if US targets China today, it could be them next.
> I'm kind of wishing this trend will push investment in something open like RISC V.

This is indeed happening.

The challenge is on the GPU side. Now is they could finalize the V extension people could start using that for graphics.
There has to be startups already doing just that
Yep, the United States government gaining the ability to directly block the licensing of ARM reference designs from companies like Huawei's HiSilicon (and the fabless chip designers such as Rockchip) is a VERY big deal. It's a very different situation to the US government having to pressure Japan's SoftBank, UK's ARM Holdings (or their respective governments).
(comment deleted)
Apple seems to be against use of nvidia GPUs in their products - would they potentially abandon using ARM if it comes under nvidia ownership?
What's the available alternative for Mobile? X86 has never matched ARM on power efficiency vs. compute trade-off for mobile
I think the commenter is referring to Apple's switch to ARM for laptops.
I'm not an expert, but I thought Apple's reason for ditching Nvidia was speculated to be that their ecosystem was closed and integrating Nvidia slowed down development and lead to failures. I don't think there's an absolute opposition to using Nvidia owned stuff, especially for ARM where Apple licenses IP and custom designs the hardware on their timeline.
Apple's customers don't care about NVidia -- they don't do serious computation, scientific computing. Our Hollywood clients all use PC-architecture machines with NVidia for video editing and rendering. The writers, etc, use Macs.

So Apple can save the cost and get higher profit margins by not having powerful GPUs in their products.

I know a lot of VFX and animation people that do serious rendering / ray-tracing computation and do wish they could use nVidia GPUs in their Macs. It's worth considering how, unlike writers, rendering needs machines not just for all the animators, but for the render farms too.
Apple typically has forward-looking supply chain contracts. I imagine ARM is under contract for the licensing for long enough that Apple will build out their own design team.

Already, TSMC is the manufacturer for the cores, so Apple really doesn't have too much of a dependency on ARM in the short term.

I think what we'll see over the next 3-5 years is a divergence from the ARM-licensed design to an in-house-designed architecture: "Apple Silicon II"

Apple already has a perpetual license for the Arm ISA (analogous to the api). The actual chip design has been Apple customer for several years now. Apple does not get their chip designs from Arm.
So in that case they really need nothing from ARM at this point
Of course not.

They might be against nVidia GPUs. They won't throw their investment in Apple Silicon out just because of that.

what if nvidia increases the amount of money apple has to pay for arm licensing
(comment deleted)
Apple is not fundamentally against the use of nVidia products. nVidia refused to implement the Metal APIs knowing full well that in a few years they'd be replaced by Apple's own silicon. Why spend a 1-2 years implementing adding a feature that will be legacy 1-2 years after it ships?
> Apple is not fundamentally against the use of nVidia products.

Are you sure about that? I don't know this story about Metal, but I believe Apple has refused to certify an nVidia driver on Mac since like 2018 (I think), effectively cutting Apple users off from using nVidia products on Apple platforms.

Apple is currently against NVIDIA GPUs simply because they got screwed over with NVIDIA GPUs before in terms of hardware reliability and such, which prompted them to switch to AMD for GPUs. It doesn't have much to do with NVIDIA as a company and more to do with their hardware.

But when it comes to ARM chips, Apple designs their own ARM-compliant chips, so NVIDIA owning ARM would do nothing in that aspect imo. Especially since iPhone/iPad chips have been ARM-based for the past decade, so I don't think that Apple really has a good option or reason to switch away from ARM at this point.

> Apple is currently against NVIDIA GPUs simply because they got screwed over with NVIDIA GPUs before in terms of hardware reliability and such

That's the oft-mentioned and certainly plausible reason, but it's not a matter of fact that it's THE reason, is it?

In the absence of any other known reason, this seems to be the most likely explanation.

But yeah, you are correct, there is no factual officially stated reason, and I wouldn't expect Apple to publicly bring it out ever either.

The argument against this reason is that Apple used Nvidia GPUs in their MacBooks from 2012 to 2014(?). This was years after the broken GPU time.
Apple has an ARM architecture license from the very early days. I don't know if it's perpetual or what the terms are. b Apple doesn't use any stock ARM cores and already develops their own compiler, libraries, OS so how much do they rely on ARM, Inc?
My theory on why Apple prefers AMD follows from Nvidia's main advantage over AMD being software, both their drivers and CUDA.

Apple needs/wants to be in full control of their driver stack. They don't necessarily want to write it themselves, but they want tighter control than Nvidia is willing to give them. AMD on the other hand is much happier to offload responsibility and accept help.

I don't think this set of circumstances would come up for an ARM processor.

Nvidia-provided macOS drivers had QC issues (major bugs) too, some of which never got fixed, which probably didn't help matters.
Apple has a perpetual license no? So wouldn’t matter too much on leverage.
Apple had a long patent and design mark infringement fight with Samsung all while using Samsung fabs for their SOCs, flash, and LCD panels.

Nothing personal, just business.

Could hopefully get Google to invest millions in RISC-v
I wonder if it would be possible for a consortium of companies like Apple, Microsoft, and Google to swoop in and outbid Nvidia? All of them rely on customization agreements with ARM and Nvidia, being a chip maker, would be competition. And a consortium like that would allow the consortium members to keep their changes to themselves so whatever Qualcomm and Microsoft are doing with the SQ1 or Apple with their Silicon stuff wouldn't have to be shared will all consortium members -- or a competing chip maker like Nvidia.
So much like a shared patent pool, which in effect does partially describe ARM.

But when those companies can get access to those patents for more accountant friendly costs without balancing out risk/asset/management aspects of owning part of that slice, then the motivation, corporate culture that would be needed to drive such an investment is just not there.

That's why having at least three "heavies" as part of the consortium will enforce good behavior or at least keep it from getting out of control. Well-armed (legal teams) companies are polite companies... (except Oracle, but they are mainly a litigation firm that also sells databases).
True, yet the incentive is just not there business wise over what they have now and be a hard sell to the board as what advantages does it give them?

It actually makes more sense for some governments to buy ARM than many companies, so at least we can be thankful (so far) that avenue has not transpired. Might even be best overall if ARM was partially IPO'd. If SOftbank sold 50% stake via IPO, I dare say that way of selling the company would yield the best and also quick return and best of both Worlds and that option I'd still not rule out. Indeed, mooted interest from many large companies sniffing as a prelude to an IPO may not be unheard of.

I too share your aversion towards Oracle's business practices.

I think you'd have significant regulatory back-pressure preventing Microsoft, Google, and Apple - between them they effectively own all the operating systems and app stores of the world. There's a real problem here that will prevent competition from outside (QCOM & Samsung are two massive ARM chip designers for mobile, there's a bunch more in the embedded space). It's important to look at the risks and not just the upsides as there are both.
I think it would make even more sense for such a consortium to buyout AMD at this point.
Which would make the x86 licensing deal void IIRC.
Nvidia might a chip maker, but they aren't really a competing chip maker. The only real competitor to Nvidia is AMD, which does not use ARM IP in any significant amount that I know of.

Nvidia might be tempted to block ARM customers if they had competing designs to push them to, but they don't. So they would be sacrificing revenue without any other sales to make up for it. It doesn't seem like a good idea to me.

This would be a different story if a company like Apple were buying ARM. They could definitely benefit by gouging Samsung and Google on license fees but they've got to balance that with the chance of getting fined for anti-competitive practices.

Apple has a perpetual license on the ARM instruction set, so they just pay a royalty per chip. Apple owning ARM would be a massive problem for regulators, as they'd be the sole source of chips to their competition.

MS and Google don't make chips, and MS doesn't even license any ARM tech (I don't think, they just make software and buy CPUs, they don't make any). They're a bad fit. Google's dipping their toes in it, but I don't think they're doing fully custom silicon. Most companies buy the rights and layouts, and tweak those layouts.

NVidia makes chips, but they don't make many mobile chips, and virtually no chips for phones compared to Samsung and Qualcomm. They're a better fit.

(comment deleted)
The chipset in the Nintendo Switch is a mobile chipset by NVidia.
Yes yes, there are outliers which is why I didn't make a blanket statement NVidia doesn't make anything mobile.
Google has been making TPUs and similar high end chips, in addition to their Titan chips.

I wouldn't be surprised if they make their own networking chips.

Apple has never shown much interested in a lot of B2B type licensing companies that aren't 100% apple, and potentially owning something their competitors need and all the legal complications that brings.

Meanwhile Apple's home grown chip initiatives are mature, not sure they need much from ARM.

Interesting, Nvidia has been using their own RISCV designs in their GPUs. Perhaps they are more interested for their automotive line?
I think it's for the datacenter. They have already purchased Mellanox. In many HPC applications, having fast GPUs and fast interconnects is 99% of what you need. A reasonably competent CPU architecture would round this out, and I don't think RISC-V is there yet.

ARM have also made some HPC plays, e.g. buying Allinea (probably the best supercomputer debugging tool).

So basically, it will be two companies which own both the CPU and GPU stack (AMD/ATI and Nvidia/ARM) and intel will just sort of end up at the wayside. Not really what I expected.
That could be in Intel’s favor if this becomes an example of “deworsification”. Something to be said about focusing purely on one use-case.
Intel's been moving in the direction of launching a discrete GPU line for years: https://www.theverge.com/2020/1/9/21058422/intel-dg1-discret...

And I'd hardly call Intel focused on one use-case. They seem to have their hands in all sorts of random-as-heck product lines like IoT devices and such, most of which rarely see much long-term support.

I think they mean that Nvidia / ARM are currently each focusing on what use case and a merger could make them worse at both.
The random stuff at Intel are usually little experiments to see how complimentary tech drives Xeon sales. When they pitched IoT stuff, the real "sale" was to get Intel-based edge devices in place to herd the IoT. And to get "intel" about what customers are planning in the place around opportunities/threats like 5G.

I think that effort in particular (about 5 years ago for me) was a warning sign about Intel -- they didn't internalize that your smart water meter would have a chip beefy enough very soon to connect to a cellular or other long range network and just hit a cloud endpoint.

FWIW, Intel has been pushing towards launching a dedicated discrete GPU platform. If/when they recover, this would place us in having three distinct CPU/GPU companies.
From what we know so far (https://www.tomshardware.com/news/intel-xe-graphics-all-we-k...), it will be a while before Intel competes in the GPU space. The first offering of Xe graphics (still not out yet) will probably not be competitive with cards that AMD and Nvidia released over a year ago.

If Intel can survive their current CPU manufacturing issues, manage to innovate on their design again, and manage to improve the Xe design in a couple generations, they might be in a good position in several years. I (as a layman) give them a 50/50 shot at recovering or just abandoning the desktop CPU and GPU market.

I wonder, why Intel abandoned the mobile market? I think only a single smartphone was released using an Intel CPU.
I had one model from Asus, battery life was terrible and if you were doing anything remotely intensive it doubled up as a nice hand warmer...

I know that these probably are solvable problems, but they left a pretty bad taste...

I think radio tech is an issue in the mobile space. It is heavily patented and the big players who can integrate this technology into their chips can offer much more energy efficient solutions.
Intel had that in-house too... it just also kinda sucked too...

That is the thing with a lot of these side projects Intel is always working on. It would be great if they actually delivered good products, but they often spend billions acquiring these companies and developing these products only to turn out one or two broken products and then dump the whole project.

I think this time is different with Xe, but I can't blame anyone for looking at the past history and being dubious that Intel is in it for the long haul.

The first TAG Heuer android wear device ran an atom chip. Fun times with the power optimization, that was.
Not enough profit for their liking was the reason.
They had a bad product that they had to sell at below cost to get market share. The x86 tax is pretty small for a big out of order desktop core but it's much more real at Atom's scale.
> The first offering of Xe graphics (still not out yet) will probably not be competitive with cards that AMD and Nvidia released over a year ago.

Being a only a year behind market leaders with your first product actually seems pretty impressive to me. Especially if that's at the (unreasonably priced) top of the line, and they have something competitive in the higher volume but less sexy down market segments.

Intel's i860 was released in 1986, that evolved to the i740 in 1996, and later on to the KNC, KNLs, Xeon Phis, etc.

The >= KNC products have all been "one generation behind" the competition. When the Intel Xe is released, Intel will have been trying to enter this market for about 30 years.

This market is more important now than ever before. I hope that they keep pushing and do not axe the Intel Xe after the first couple of generations only to realize 10 years later that they want to try again.

Not their first try on that front, and Intel push for new product branches has not exactly gone well the past decade or so.
If I was Intel, I would be going straight for the TPU market. GPU have a bunch of legacy from the G=Graphics legacy. The real money maker is not likely to be gamers (although it has been healthy enough market). The future of those vector processing monsters is going to be ML (and maybe crypto). This is the difference between attempting to leapfrog compared to trying to catch up.
This is what they tried to do with Nervana and are trying again to do with Habana.
> The future of those vector processing monsters is going to be ML (and maybe crypto)

That's a heavy bet on ML and crypto(-currency? -graphy?). Has ML, so far, really made any industry-changing inroads in any industry? I'm not entirely discounting the value of ML or crypto, just questioning the premature hype train that exists in tech circles (especially HN).

> Has ML, so far, really made any industry-changing inroads in any industry?

Does the tech industry not count, or are you only considering industries that are typically slower moving?

The tech industry claims it applies machine learning all over the place, but I doubt it actually moves the needle much.
Is that opinion based on anything firmer than a pessimistic outlook?

Lots of people jump on any trend, it doesn't mean the hype is unjustified or that nobody is "moving the needle" with that trend. Recommendations (e-commerce, advertising, music, etc) have been pretty revolutionized by it.

> Recommendations (e-commerce, advertising, music, etc) have been pretty revolutionized by it.

I'd like to learn more about how they've been revolutionized by it. Any sources you could share?

As a personal anecdote, the quality of the recommendations I receive across the board has been roughly inversely proportional to the level hype around ML in the tech press and academia.

C.f. Youtube's, Amazon and Netflix (products that bet BIG on recommendations) being incapable of recommending compelling material.

ML contributes to a significant fraction of revenue at three of the world's largest companies (Amazon, Google, Facebook - largely through recommendations and ad ranking). It also drives numerous features that other tech companies build into their products to stay competitive (think FaceID on iPhone). Hard to argue that it doesn't move the needle...
(comment deleted)
>That's a heavy bet on ML and crypto

Well, yes that is the point. My theory is that the gaming market for GPUs is well understood. I don't think there are any lurking surprises on the number of new gamers buying high-end PCs (or mobile devices with hefty graphics capabilities) in the foreseeable future.

However, if one or more of the multitude of new start-ups entering the ML and crypto (-currency) space end up being the next Amazon/Google/Facebook then that would be both unforeseeable and unbelievably transformative. Maybe it won't happen (that is the risk) but my intuition suggests something is going to come out of that work.

I mean, it didn't work out for Sony when they threw a bunch of SPUs in the PS3. They went back to a traditional design for their next two consoles. So not every risk pans out!

> crypto(-currency? -graphy?)

TPUs are massively parallel Float16 engines - not really applicable to anything outside of ML.

> Has ML, so far, really made any industry-changing inroads in any industry?

It is (IIRC) a pretty fundamental part of self driving tech. I honestly think this is what drives a lot of Nvidia's valuation.

nvidia's largest revenue driver, gaming, made 1.4B dollars last year (up 56% YoY). nvidia's second largest, "data center" (AI) made 968M (up 43% YoY). Other revenue was 661M. Up to you if nvidia's second largest revenue center, of nearly a billion/year is "industry changing"
A ton. Look at the nearest device around you, chances are it runs Siri, Alexa, Cortana, or Google voice assistant. This will only grow.

Same with machine vision. It's going to be everywhere — not just self-driving trucks (which, unlike cars, are going yo be big soon), but also security devices, warehouse automation, etc.

All this is normally run on vector / tensor processors, both in huge datacenters and on local beefy devices (where a stock built-in GPU alongside ARM cores is not cutting it).

This is a growing market with.a lot of potential.

Licensing CUDA could be quite a hassle, though. OpenCL is open but less widely used.

> The future of those vector processing monsters is going to be ML (and maybe crypto).

Hopefully some of those cryptocurrencies (until they get proof-of-stake fully worked out) move to memory-hard proof-of-work using Curve25519, Ring Learning With Errors (New Hope), and ChaCha20-Poly1309, so cryptocurrency accelerators can pull double-duty as quantum-resistant TLS accelerators.

I'm not necessarily meaning dedicated instructions, but things like vectorized add, xor, and shift/rotate instructions, at least 53 bit x 53 bit -> 106 bit integer multiplies (more likely 64 x 64 -> 128), and other somewhat generic operations that tend to be useful in modern cryptography.

They have. Their first purchase (Nervana) hasn’t worked out for them so they are now working through their purchase of the more conventional Habana.
The one thing I don't get is, there are a lot of machines out there that would gain a lot from specialized search hardware (think about Prolog acceleration engines, but lower level). For a start, every database server (SQL or NoSQL) would benefit.

It is also hardware that is similar to ML acceleration, it needs better integer and boolean (algebra, not branching) support, and has a stronger focus on memory access (that ML acceleration also needs, but gains less from). So how comes nobody even speak about this?

I don't understand how database servers would benefit. You would have to add the search hardware directly to DRAM for any meaningful gains.
You would need large memory bandwidth and a good set of cache pre-population heuristics (putting it directly on the memory is a way to get the bandwidth).

ML would benefit from both too, as would highly complex graphics and physics simulation. The cache pre-population is probably at odds with low latency graphics.

Intel did go for the tpu market. It was called the nervana chip, and they cancelled it.
Isn't this a monopoly? doesn't seem very legal. If the future is so dependent on GPU compute, and there are only 2 companies (AMD/NVIDIA) making them.... isn't it insane to only have 2 companies the entire world depends upon?

Edit: ok Duopoly, but still... kinda insane that only 2 companies in the world do it.

It's a duopoly, and unless you've got evidence that AMD and Nvidia are colluding to hurt potential competitors there's nothing illegal about it.
It’s really hard to call a duopoly much better than a monopoly. Sure you can probably identify points, but neither outcome is anywhere near efficient. I don’t know that it should necessarily be illegal. But it is market concentration.
A duopoly is absolutely much better than a monopoly. You've got competition pushing you improve.
but by definition isn't a duopoly not competing with each other? I might be wrong, but I remember it being about the two companies making deals with each other to keep both prices high and not innovate or improve, like ISPs in the US
No, that is not a characteristic of a duopoly and in many cases is illegal.
Oh it’s illegal so they won’t do it or have a zillion ways to work around it while being technically legal.
Anticonsumer practices really need a test of the impact on consumers, similar to how hiring practices are tested. There is a give and take in most broadband suppliers, for example, where there is no written agreement to not compete but they don't compete.
Part of it is the software world's fault as well. Almost the entire open source machine learning ecosystem is written for NVIDIA's CUDA and nothing else.

And almost all competent non-CUDA platforms (e.g. Google TPUs, Tesla's secret in-house hardware) haven't been open-sourced, or even sold to consumers, which further enables the NVIDIA monopoly.

While ARM can be great, I don’t think anyone is entirely writing off x86. Intel owns a lot of fab. And while I don’t get as hyped on it as other people, Intel going hard into RISCV could change the game for ARM (if it was handled in non-typical-Intel way)
(comment deleted)
I'm not writing off x86, but I think AMD is going to define the future of x86 (like the did before), instead of Intel.
> I don’t think anyone is entirely writing off x86.

Apple seems to be.

That’s fair.

I was about to write how that’s a special case, but then I remembered that Microsoft is also making their own chips for Surface products.

But Intel has already threaten MS with a lawsuit if it tries to emulate x86 for its ARM products. Meanwhile when Apple switched from PowerPC to x86 part of the agreement was a share of patents which gave Intel access to the multimedia extensions of the PowerPC (AltiVec?). If this "share" went both ways then Apple might be able to provide x86 translation on their SOC. NOTE: there is no evidence of this anywhere other than what MS was blocked when working on Windows 10 for ARM and the Apple-Intel agreements.

If this does happen I think Intel will not sweat it because it will be only Apple. Apple has no interest in selling CPUs. They want to be able to make severe changes (cut the fat) between revisions and not have people crying about having to update their architecture to support it.

IANAL, but it's been argued here that the x86_64 patents will expire soon (the specification was available in 2000, first processor in 2003). Probably neither Apple or MS will be blocked; MS had a problem earlier since it decided to release while the patents were still in force, rather than wait.
Nvidia hardware is great but my biggest concern is that this would eventually spell the end for Raspberry Pis and other good things that are ARM. That would not be cool.
Can you expand on that concern? Rpi uses a Broadcom soc.. so it isn’t like they deal with arm directly.
Designs of future ARM versions could end up being skewed toward NVIDIA's needs or worse, designed to or information withheld in ways that kill other embedded products and monopolize the embedded market.

They could theoretically simply stop licensing ARM to Broadcom, although that might invoke some anti-trust suits.

Broadcom is Acorn, there is a suggestion in another comment that Apple has a perpetual licence to ARM, maybe Broadcom has as well.
Do you not know Intel is close to releasing their own GPUs or do you just think they will just fail at it?

Even with the acquisition of ARM, I don't see Nvidia any better off than Intel at this moment as far as CPU/GPU stack goes. Frankly, I would think AMD would be the one to end up by the wayside since they still are weak on the software side.

I think there's a good chance they'll fail. This is true of any new venture, so it's a bit lame to say, but there are reasons:

Right now I have a fairly decent GPU in my Macbook which I've hardly used. Very little supports it, because it's not nVidia. I can't use it for AI training, for example. Sure, it might work ok for some games, but Macbooks aren't really for gaming, and nVidia has captured that market nicely anyway.

Things can change; maybe Intel's software stack will be incredible. I don't know. But they have quite a hill to climb before they reach that summit.

Intel has good record to provide software, contribute to OSS, support for developers to support hardware than AMD, so possibly they can do better than RADEON on some world if they can provide great hardware.
There's ROCm[1] though. It's just almost every ML platform blindly bent to the NVIDIA vendor lock-in. CUDA is a disaster, like DirectX was back in time. One day it will go, hopefully soon enough.

[1] https://github.com/RadeonOpenCompute/ROCm

There's a huge market for mobile devices that aren't Apple and are fast. Perhaps this would offer better SoC prospects than whatever slop Qualcomm is dishing out.
Where is this “huge” non Apple mobile market? The mobile market outside of Apple is a commoditized race to the bottom. The average selling price of an Android phone is around $270.

https://www.statista.com/statistics/951537/worldwide-average...

Hardware always gets cheaper and better. Why do you think Apple is so aggressively pushing services?
Yet and still Apple has been charging a premium over competitors for over 40 years....
They segment the market and de-empathize cheap stuff. Even so, as many iPhone 7s ($200) were activated as iPhone 11s as recently as January.
They will fail. If they release anything less than a card as fast as a standard nvidia card and with as good driver support, they have absolutely no chance.

nvidia has a huge market, people buy their stuff because it's fast and there's software support for everything from AAA games to scientific/engineering modelling to ML.

intel took far, far too long to come up wiht a viable graphics ecosystem to ever be successful.

Most GPU sales are for $200-$400 cards. AMD has always optimized for that instead of having the fastest card outright.
Doesn’t intel ship the most GPUs? Just not discrete. Their iGPUs are in so many low end systems compared to any systems with discrete graphics cards
>Do you not know Intel is close to releasing their own GPUs or do you just think they will just fail at it?

Given this is about their third attempt at releasing a high performance gpu, I think skepticism is warranted until they’re actually selling something to the general public.

Maybe I'm jaded, but my current hope is they will fail at a discreet GPU, then dump that tech into making their onboard much better and using that as a selling point for the CPUs, thus helping everyone in the long run.
I wouldn't count Intel out yet. Sure, things don't look that great, but same was true of AMD/ATI until recently...
Is this comparable to that situation? AMD & Intel were building chips basically to the same standard. ARM/Nvidia vs. Intel would be a lot more asymmetric.
> I wouldn't count Intel out yet. Sure, things don't look that great

I always get a kick out of the sentiment toward Intel on HN.

Intel is booming financially. Things have never been better for them in that respect. They have every opportunity to fix their mess.

Intel has eight times (!) the operating income of Nvidia, with a smaller market cap.

Intel is one of the world's most profitable corporations. $26 billion in operating income the last four quarters. Their margins are extreme. Their sales are at an all-time high. Their latest quarterly report was stellar.

In just 2 1/2 years Intel has added a business the size of Nvidia and AMD combined.

If they can't utilize their current profit-printing position to recover, then they certainly deserve their tombstone. Nobody has ever had an easier opportunity to find their footing.

Right, that's kinda my point, maybe not strongly enough stated. The "don't look that great" is "they'll probably have to buy fab resources from someone else" to keep up perf-wise (like their competitors, who are fabless), not "they're going bankrupt anytime soon".
Intel has a habit of putting snatching defeat from the jaws of victory.

Their decades of more or less monopoly status has made them complacent, their revenue is high still because of this inertia built into the market that simply will not vanish.

The datacenter for example is still dominated by Xeon not because people like Xeon over Epyc but because there is not a easy migration path between the 2 platforms. If I was building a whole new server farm with all new VM's I would choose Epyc all day... but if I need to upgrade hosts in an existing farm with no down time well that will need to be Xeon then...

When it comes to desktops/laptops though, Lenovo's AMD line is attractive and suffers no such problems

(comment deleted)
This sounds very similar to the situation Nokia was in around 2007: - Nokia was booming financial. Things had never been better for them in that respect. - Nokia had XX times (!) the operating income of Apple's mobile business. - Nokia was one of the world's most profitable corporations.

And, yet, the writing was on the wall. Nokia was doomed once the smartphone era came. That's where Intel is today: AMD crushes them on the high-end general purpose CPUs. ARM crushes them on I/O performance and the low-end for general purpose CPUs. GPUs crush Intel in the middle, for special-purpose (mainly single-precision floating point) computing.

Right now, large portions of new computer sales, and an even larger portion of the high-margin cpu sales, come from cloud computing. AMD and ARM are stealing huge market share from Intel on that front. I don't see that momentum changing any time soon.

There's a reason that Intel has 8x the operating income of NVidia while having a smaller market cap. It's not because of where they are currently--it's where they are going. Stock market valuations are forward-looking, and the future doesn't look so bright for Intel.

Stock market valuations are forward-looking, but they aren't always predicting the right future.

Personally I won't be betting on or against Intel - it wouldn't shock me if they follow the Nokia route, it also wouldn't shock me if they come out with a new generation that puts them back on top within the next few years.

> ARM crushes them on I/O performance

This sounds difficult to believe. Do you know a benchmark that shows this?

(comment deleted)
They have created a new type of memory that is 10-100x more performant than flash, and its already popular in high-performance databases. 3dx point

That could be a huge market with no cempetutors in sight

Nokia was fine - just switch flagship from Symbian to Android, continue feature phones and Maemo. After so many years and under different manufacturer brand is still alive.
Intel is booming financially. Things have never been better for them in that respect. They have every opportunity to fix their mess.

So was RIM in 2010. Profits are a trailing indicator. The PC market is really small compared to the mobile market and declining. While Apple only has 10% of the overall market, it has a much higher percentage of high end personal computers and Intel is about to lose Apple as a customer.

PCs also are having longer refresh cycles. What does “recovery” look like? PC sells going up? That’s not going to happen.

They still have the server market while that is probably growing, Amazon is pushing its own ARM processors hard and MS and Google can’t be too far behind.

I find it humorous too, but I think it's easily explained: we get a constant stream stories of the plucky underdogs with their fancy engineering achievements. HN loves both industry disruption and engineering achievements, so it's a sort of self-reinforcing reality distortion field. See Tesla for a very similar sort of story.

The innovators and underdogs are always great to see, and they fuel our collective imagination, so it's no surprise that they dominate the HN front page. Of course, that mind-share dominance is in stark contrast to the well-entrenched money-printing machines they're trying to disrupt, who are happy to keep dominating their respective industries year after year instead.

> Their sales are at an all-time high. Their latest quarterly report was stellar.

Past performance. In spite of their earnings their stock plummeted on the 7nm news. It’s not just HN that is bearish.

This is very true. They got this profitable though by gutting every long term investment in new forays: they sold off their ARM business, their modem business, they never bothered to make a serious GPU or mobile chip, etc.

The margins on those big Xeon chips have been so good that they ditched everything else, and painted themselves into a corner, sitting by the sidelines for the past 20 years as new markets emerged.

Intel has something like 60% of the PC GPU market because of their built-in graphics.
Yes, but not the high-end GPU market, which is what Intel has been trying to break into for quite some time now.
ARM is not in the high end GPU market either.
No, but Nvidia is. So if Nvidia bought ARM, they could be a CPU and GPU maker.
Nvidia has had modern ARM chips (the Tegra series among other) for at least a decade now. Any company can license ARM tech.
The top level comment was about companies owning a CPU and GPU stack. NVIDIA licenses ARM IP now, if they bought ARM they would actually own the designs, putting them on the level of AMD and Intel where they would have greater control over the technology.

Just licensing the designs puts a company on a lower chipmaking tier with Qualcomm, Samsung, Apple, Huawei, etc.

> NVIDIA licenses ARM IP now, if they bought ARM they would actually own the designs

NVIDIA Carmel is NVIDIA's own Arm v8.2-A design [0]. They only licence the architecture. The core is quite interesting, as it's doing dynamic recompilation for an underlying VLIW architecture.

However, NVIDIA Orin[1] (followup for Tegra Xavier) will use a Cortex-A78 core[2] (formerly Hercules) licenced from ARM.

[0] https://en.wikipedia.org/wiki/Project_Denver

[1] https://nvidianews.nvidia.com/news/nvidia-introduces-drive-a....

[2] https://en.wikichip.org/wiki/arm_holdings/microarchitectures...

Heh... ever since the days of the chips and technology acquisition, if not sooner, they vacillate between wanting to be in the graphics business and not being in the graphics business...
My impression was they were intentionally not competing with Nvidia at one point?
From the outside it sounds more like Intel has been infighting about this for 20+ years...

The original i740 was theoretically a capable card; although fairly hampered by being forced to use Main Memory for Textures. Intel eventually backed down from the graphics market back then, and instead continued to use the 740 as a basis for the integrated graphics in the i810/815 chipsets.

But, as GPUs became closer to what we saw as real GPUs, Intel continued to press on with the idea that keeping things done in the CPU was better for them (i.e. encouraging upgrades to higher end CPUs vs selling more lower margin graphics cards.)

You saw a similar pattern with the 845/855/865: Shaders were all done in software (Hey, it finally almost justified Netburst, right? ;)

And this pattern seems to continue with various forms of infighting between groups up to this day.

The other Consistent problem they have had is driver compatibility/capability.

Also, the i740 drivers were really bad. I had one and I remember all kind of bug and graphical glitchs on games that worked fine on a GeFoce 2 MX that I got latter.
Aren't AMD's APU's a better product for that market and will replace intel sooner or later?
One difference is that AMD only offers integrated graphics with weak CPUs, while Intel, unless something's changed recently, offers them on all their consumer models.
Intel has started offering "F" SKU processors without integrated graphics only very recently with 9th and 10th gen Core CPU's.

But yes otherwise, nearly every Intel CPU has integrated graphics whereas only a few select AMD CPU's have integrated graphics (and AMD brands them as APU's not CPU's).

I'd be okay with only a few select CPUs, if even one of them was a reasonably powerful one. Instead, it's only the bottom of the barrel CPUs performance-wise.

It seems that is changing somewhat with the 4000-series APUs, but guess what, those are only going to be sold to OEMs, not individuals.

It's all rather frustrating, since I'm still on an i7-4770k and wouldn't mind an upgrade.

So your situation is you want to DIY build a desktop with an integrated CPU+GPU, with a significant upgrade from a 4770k.

4770k passmark: 7042 https://www.cpubenchmark.net/cpu.php?cpu=Intel+Core+i7-4770K...

Ryzen 5 3400G passmark: 9421 https://www.cpubenchmark.net/cpu.php?cpu=AMD+Ryzen+5+3400G&i...

Intel Graphics 4600 (in 4770k): 649 https://www.videocardbenchmark.net/gpu.php?gpu=Intel+HD+4600...

RX Vega 11 (in 3400G): 2106 https://www.videocardbenchmark.net/gpu.php?gpu=Radeon+RX+Veg...

So relative to 4770k, 3400G has a 33% faster CPU, and 3.25x the GPU.

> 3400G has a 33% faster CPU

Right. That's not a significant upgrade IMO. I'm not even sure it would be worth it if it merely required a CPU swap. Since it actually requires a new motherboard, it's not even close to worth it.

A real upgrade would be to a 3900X (passmark: 32861), or at the very last, a 3600 (passmark: 17828). But those require a discrete GPU.

The 4700G looks like it would more or less suffice (passmark: unknown, ~18k?) , but it won't be sold to individuals, only OEMs.

> 3.25x the GPU.

As far as I can tell, I've never run into any limits of the 4770K's iGPU, so I don't think this matters. Running dual 1920x1200 monitors.

It would also let you upgrade from DDR3 to DDR4 and ~double your memory bandwidth. But if you wait another year or two, you could jump right to a DDR5 system :)
Good point. Plus, of course, NVMe.

Maybe I should review AMD's GPU offerings again. Do you happen to know anything about this? Last time I was looking for (fanless + dirt cheap + dual display), and couldn't find anything that fit all 3. However... I didn't ask the question, does the fan run all the time, or only under heavy load?

Also, with lots of games reportedly working on Linux these days, maybe I should replace "dirt cheap" with "reasonably cheap."

(comment deleted)
AMD bet on APUs more than a decade ago, hoping to grab the market with better GPUs than Intel, and have been offering a better product about as long.

But the reality is that vast majority of business desktops and laptop don't need anything better than what Intel offers.

If AMD gains share in that segment (and it looks like they are), it will IMO be because of finally having a better CPU, not because of a better GPU.

Only if Nvidia stop licencing ARM to other companies like Apple, Qualcomm and … AMD, right?
As a founder of ARM, Apple is grandfathered into a "perpetual architecture license," clear sailing unless nVidia deprecated ARM in favor some something completely new (as I understand it), which seems unlikely to say the least.
> As a founder of ARM, Apple ...

Apple was a founding member of AIM (Apple, IBM, Motorola) for PowerPC. (As rumour has it, after DEC refused to go for a higher-volume lower-margin Alpha AXP derivative when Apple came knocking for an m68k replacement, Apple then asked IBM for a higher-volume lower-margin POWER derivative, leading to AIM and PowerPC.) As far as I know, only Acorn had a role in founding/spinning off ARM.

On the other hand, I would be very surprised is Apple wasn't smart enough to get a very-long-term/perpetual license on the ARM instruction set before investing heavily in custom core design.

https://en.wikipedia.org/wiki/ARM_architecture

"In the late 1980s, Apple Computer and VLSI Technology started working with Acorn on newer versions of the Arm core. In 1990, Acorn spun off the design team into a new company named Advanced RISC Machines Ltd."

https://www.latimes.com/archives/la-xpm-1990-11-28-fi-4993-s...

" Apple has invested about $3 million (roughly 1.5 million pounds) for a 30% interest in the company, dubbed Advanced Risc Machines Ltd. (ARM), but the exact ownership stake of VLSI and Acorn was not disclosed. "

Today I learned! Thanks, and upvoted.
They needed it for the Newton!
Yes and they used the StrongARM by DEC (who later sold it to Intel as part of a lawsuit settlement)

https://en.wikipedia.org/wiki/StrongARM

IIRC they used StrongARM for the last models, the Apple MessagePad 2000 and 2100. (I bought a used MP2K after the Newton was discontinued that I used in high school. Fun little device!)
Apple was part of the change from "Acorn RISC Machine" to "Advanced RISC Machines", so basically a founder of modern ARM. AIM was completely separate and later.

Apple and a bunch of others also have a "Arm architectural licence" which IIRC cannot be taken back.

Apple also fully controls the hardware and software stack for both CPU and GPU.
(comment deleted)
SoftBank purchased ARM for $32B in 2016. If this figure is correct it looks terrible for SoftBank. Another one in a string of bad investments.
Depends how much they've milked it in the meantime.
It’s growth has slowed and profits are down. I doubt they will get $20B for it.
I really hope this doesn’t happen. Nvidia could hike prices for future designs, making devices more expensive as a result.
This will be killed or severely hampered by antitrust agencies. Definitely by the EU, maybe the US as well if there's anyone decent left there.
Why? From where I sit there will still be plenty of (or at least no less) competition in the CPU/GPU space--especially considering the fact that ARM itself is not a semiconductor manufacturer.
Why definitely? What if Nvidia is looking to get out of producing GPU hardware and move to an ARM model?
Antitrust nowadays is just a buzzword politicians throw around every couple years in reference to household names like FAANG in order to get reelected. Unless your average grandma knows the name Nvidia, don't count on those in power to even pretend to do something about this.
AMD ATI deal went through , if this got blocked , isn’t it antitrust other way around ? You are not letting Nvidia compete with AMD?
This would be bad. Not because of the CPU business - I think RISC V will eventually make that irrelevant. Once CPUs are open source commodities, the next big thing is GPUs. This merger will eliminate a GPU maker, and one that licenses the IP at that.
Why wouldn't something similar to RISC V happen in the GPU space?
RISC-V is an ISA rather than silicon, GPUs are generally black boxes that you throw code at. There's not much to standardize around
GPUs do have ISAs. It's just that they're typically hidden behind drivers that provide a more standardized API.
Of course they have ISAs, my point is that the economics of standardization around a single ISA a la RISC-V isn't as good by virtue of the way we use GPUs on today's computers. You could make GPU-V but why would a manufacturer use it
> GPUs are generally black boxes that you throw code at.

umm... what? what does that even mean? lol

I could kind of maybe begin understand your argument from the Graphics side, as users mostly interact with it at an API level, however keep in mind that shaders are languages the same way "cpu languages" work. It's all still compiled to assembly, and there's no reason that you couldn't make an open instruction set for a GPU the same as a CPU. This is especially obvious when it comes to Compute workloads, as you're probably just writing "regular code".

Now, that said, would it be a good idea? I don't really see the benefit. A barebones GPU ISA would be too stripped back to do anything at all, and one with the specific accelerations needed to be useful will always want to be kept under wraps.

You could make a standardized GPU instruction set but why would anyone use it? We don't currently access GPUs at that level, like we do with the CPU.

It's technically possible but the economics isn't there (was my point). The cost of making a new GPU generally includes writing drivers and shader compilers anyway, so there's not much of a motivation to bother complying with a standard. It would be different if we did expose them at a lower level (i.e. if CPU were programmed with a jitted bytecode then we wouldn't see as much focus on ISA as long as the higher level semantics were preserved)

Just 'cause Nvidia might want to keep architectural access under wraps doesn't necessarily mean that everyone else is going to, or that they have to in order to maintain a competitive advantage. CPU architectures are public knowledge, because people need to write compilers for them, and there are still all sorts of other barriers to entry and patent protections that would allow maintaining competitive advantage through new architectural innovations. This smells less of a competitive risk and more of a cultural problem.

I'm reminded of the argument over low-level graphics APIs almost a decade ago. AMD had worked together with DICE to write a new API for their graphics cards called Mantle, while Nvidia was pushing "AZDO" techniques about how to get the best performance out of existing OpenGL 4. Low-level APIs were supposed to be too complicated for graphics programmers for too little benefit. Nvidia's idea was that we just needed to get developers onto the OpenGL happy path and then all the CPU overhead of the API would melt away.

Of course, AMD's idea won, and pretty much every modern graphics API (DX12, Metal, WebGPU) provides low-level abstractions similar to how the hardware actually works. Hell, SPIR-V is already halfway to being a GPU ISA. The reason why OpenGL became such a high-overhead API was specifically because of this idea of "oh no, we can't tell you how the magic works". Actually getting all the performance out of the hardware became harder and harder because you were programming for a device model that was obsolete 10 years ago. Hell, things like explicit multi-GPU were just flat-out impossible. "Here's the tools to be high performance on our hardware" will always beat out "stay on our magic compiler's happy path" any day of the week.

SPIR-V looks like a promising standardization. It can not be translated directly into silicon but it doesn't have to. Intel also essentially emulates x86 and runs RISC internally.
>Intel also essentially emulates x86 and runs RISC internally.

By that logic anything emulates its ISA because that is the definition of an ISA. An ISA is just the public interface of a processor. You are wrong about what x86 processors run internally. Several micro ops can be fused into a single complex one which is something that cannot be described with a term from the 60s. Come on, let the RISC corpse rot in peace. It's long overdue.

GPUs are also much simpler chips in comparison to CPUs.

90%+ of the core logic area (stuff that is not i/o, power, memory, or clock distribution) on the GPU are very basic matrix multipliers.

They are in essence linear algebra accelerator. Not much space for sophistication there.

All best possible arithmetic circuits, multipliers, dividers, etc. are public knowledge.

Is that last sentence provable? If so, that's an impressively-strong statement (to state that the provably-most-efficient mathematical-computation circuit designs are known).
Well, at least for reasonably big integers, it is.

There might be faster algos for super long integers, or minute implementation differences that add/subtract few kilogates.

> Well, at least for reasonably big integers, it is.

Big integer calculations are the bread and butter of GPUs now?

I've been studying and blogging about GPU compute for a while, and can confidently assert that GPUs are in fact astonishingly complicated. As evidence, I cite Volume 7 of the Intel Kaby Lake GPU programmers manual:

https://01.org/sites/default/files/documentation/intel-gfx-p...

That's almost 1000 pages, and one of 16 volumes, it just happens to be the one most relevant for programmers. If this is your idea of "simple," I'd really like to see your idea of a complex chip.

The most complex circuit on the GPU would be the thing chops the incoming command stream, and turns it into something which matrix multiplicators can work.
> The most complex circuit on the GPU would be the thing chops the incoming command stream

That's not true at all. I'd recommend reading up on current architectures, or avoid making such wild assumptions.

I get the feeling you're only really thinking about machine learning style workloads. Your statement doesn't seem to take into account scatter/gather logic for memory traffic (including combine logic for uniforms), resolution of bank conflicts, sorting logic for making blend operations have in-order semantics, the fine rasterizer (which is called the "crown jewels of the hardware graphics pipeline" in an Nvidia paper), etc. More to the point, these are all things that CPUs don't have to deal with.

Conversely, there is a lot of logic on a modern CPU to extract parallelism from a single thread, stuff like register renaming, scoreboards for out of order execution, and highly sophisticated branch prediction units. I get the feeling this is the main stuff you're talking about. But this source of complexity does not dramatically outweigh the GPU-specific complexity I cited above.

Isn't the rasteriser "simply" a piece of code running on the GPU?
No, there is hardware for it, and it makes a big difference. Ballpark 2x, but it can be more or less depending on the details of the workload (ie shader complexity).

One way to get an empirical handle on this question is to write a rasterization pipeline entirely in software and run it in GPU compute. The classic Laine and Karras paper does exactly that:

https://research.nvidia.com/publication/high-performance-sof...

An intriguing thought experiment is to imagine a stripped-down, highly simplified GPU that is much more of a highly parallel CPU than a traditional graphics architecture. This is, to some extent, what Tim Sweeney was talking about (11 years ago now!) in his provocative talk "The end of the GPU roadmap". My personal sense is that such a thing would indeed be possible but would be a performance regression on the order of 2x, which would not fly in today's competitive world. But if one were trying to spin up a GPU effort from scratch (say, motivated by national independence more than cost/performance competitiveness), it would be an interesting place to start.

Intel will ship larrabee any day now... :)
The host interface has to be one of the simplest parts of the system, and I mean no disrespect to the fine engineers who work on that. Even the various internal task schedulers look more complex to me.

If you don't have insider's knowledge of how these things are made, I suggest using less certain language.

I completely disagree with this comment.

Just because a big part of the chip are the shading units it doesn't mean it's simple or there's no space for sophistication. Have even you been following the recent advancements in recent GPUs?

There is a lot of space for absolutely everything to improve. Especially now that Ray Tracing is a possibility and it uses the GPU in a very different way compared to old rasterization. Expect to see a whole lot of new instructions in the next years.

> 90%+ of the core logic area (stuff that is not i/o, power, memory, or clock distribution) on the GPU are very basic matrix multipliers. >All best possible arithmetic circuits, multipliers, dividers, etc. are public knowledge.

Combine these 2 statements and most GPUs would have roughly identical performance characteristics (performance/Watt, performance/mm2, etc)

And yet, you see that both AMD and Nvidia GPUs (but especially the latter) have seen massive changes in architecture and performance.

As for the 90% number itself: look at any modern GPU die shot and you'll see that 40% is dedicated just to moving data in and out of the chip. Memory controllers, L2 caches, raster functions, geometry handling, crossbars, ...

And within the remaining 60%, there are large amounts of caches, texture units, instruction decoders etc.

The pure math portions, the ALUs, are but a small part of the whole thing.

I don't know enough about the very low level details of CPUs and GPUs to judge which ones are more complex, but in claiming that there's no space for sophistication, I can at least confidently say that I know much more than you.

> GPUs are also much simpler chips in comparison to CPUs

Funny you say that. I've never heard a CPU architect coming to the GPU world and say "Gosh, how simple is this!".

I invite you to look at a GPU ISA and see for yourself, and that is only the visible programming interface.

Judging by your nickname, I think I have reasons to listen. You are that David who writes GPU drivers?

So, what do you think is the most complex thing on an Nvidia GPU?

Matrix multipliers? As in those tensor cores that are only used by convolutional neural networks? Aren't you forgetting something? Like the entire rest of the GPU? You're looking at this from an extremely narrow machine learning focused point of view.
AMD document their ISA: https://llvm.org/docs/AMDGPUUsage.html#additional-documentat...

That's why third party open shader compilers like ACO could be made:

https://gitlab.freedesktop.org/mesa/mesa/-/tree/master/src/a...

AMD document their ISAs, but each one maps pretty much one-to-one onto a particular implementation. Compatibility and standardization are not goals.
As long as they make an open source compiler, there is at least a reference implementation to compare to.
In a GPU the ISA isn't decoupled from the architecture in the way it is for a post-Pentium Pro CPU. Having a fixed ISA that you couldn't change later when you wanted to make architectural changes would be something of a millstone to be carrying around for a GPU.
I’m curious, why is this the case for GPUs and not CPUs?
It's much more advantageous to be able to respin/redesign parts of the GPU for a new architecture since the user interface is at a much much higher level compared to a CPU. They basically only have to certify that it'll be API compatible at CUDA/OpenCL/Vulkan/OpenGL/DirectX level and no more. All of those APIs specify that the drivers are responsible for turning it into the hardware language, so every program is already re-compiled for any new hardware. This does lead to tiny rendering differences in the end (it shouldn't but it frequently does, due to bug fixes and rounding changes). So because they aren't required to keep that architectural similarity anymore, they're free to change as they need new features or come up with better designs (frequently to allow more SIMD/MIMD style stuff, and greater memory bandwidth utilization). I doubt they really change all that much between two generations, but they change enough that exact compatibility isn't really worth working at.

If you want to look at some historical examples where this wasn't quite the case, look at the old 3DFX VooDoo series. They did add features but they kept compatibility to the point where even up to a VooDoo 5 would work with software that only supported the VooDoo 1. (n.b. This is based on my memory of the era, i could be wrong). They had other business problems, but it meant that adding completely new features and changes in Glide (their API) was more difficult.

You make a HUGE assumption that is really going to happen.
Well sure, but... aren't the consequences of the possibility that it's not going to happen literally not worth discussing at all?
Agreed, this reminds of the AMD acquisition of ATI a couple decades ago. Almost killed AMD and now Nvidia is thinking about doing the same? Technically the reverse since Nvidia is the GPU company acquiring a CPU company.

At the same time, Nvidia may be trying to hedge its future as its other competitors (Intel, AMD, even Apple) all have their own CPU and GPU designs. The animosity with Apple has shown the power dynamics and the high stakes.

I believe grandparent is referring to ARM's Mali GPU.
Apple's CPU's are still ARM64 cores at heart, even if heavily modified

A thought I've been kicking about, though I fully understand it to be incredibly unlikely would be if NVIDIA would simply terminate any license Apple has to use ARM at all. The move would arguably be done out of pure spite. "payback" for the GeForce 8600M issues that cost them 200~ million and $3 billion in market cap in part due to Apple pushing for a recall.

Apple also seemingly pushed Nvidia utterly out the door, going as far as to completely block NVIDIA from providing drivers for users to use their products even in "eGPU" external enclosures on newer versions of MacOS. Even if only a minority of Apple users ever bought Nvidia cards, being completely banned from an OEM's entire lineup would likely ruffle feathers

Aren't you mixing ISA with implementation here? Or do Apple's CPUs really contain Arm's logic contrary to public info eg on Wikipedia?
You have to license the ARM ISA as well, you can't just freely implement it. IANAL, but as far as US law is concerned, I don't think you're violating copyright by cloning an ISA, but if that ISA is covered by any patents, you'd be fairly screwed without licensing. ARM definitely considers there to be enforceable patents on their ISA (just follow their aggressive assault on any open source HDL projects that are ARM compatible).

If Nvidia bought ARM and decided to find some legal way to terminate the contract with Apple out of spite, Apple would have to find another ISA for their "Apple Silicon".

If Apple are worried about that prospect, couldn't they easily outbid Nvidia to buy ARM?
Bloomberg says that SoftBank approached Apple before they approached Nvidia, but who knows.
I should have read the article before commenting! But I guess that means Apple aren't worried, either because their contracts are solid enough and/or they already have backup plans ready that are cheaper than buying ARM.

(Or, of course, Apple are making a huge mistake. But seems a bit less likely to me.)

Theoretically Apple could probably outbid Nvidia, but realistically regulators would never let Apple make that purchase so it's irrelevant that they have the cash to do so.
I'm not particularly familiar with US antitrust law specifics...

One big point that came up in the congressional hearing the other day was how Google, when buying DoubleClick, said (under oath to congress) that not only would they not merge data but that they legally couldn't if they wanted to - and then years later did just that.

Is there any way to acquire in such a way that Apple would own ARM but there'd be a complete firewall between them, with ARM having a separate board, CEO, etc. and nothing between them except the technical ownership (and any contracts between the two companies)?

I hope not, as I'm in favour of breaking up huge companies myself... but if legal firms can have a system in place for firewall of information between clients I don't see why a similar legal situation could be feasible for allowing Apple to buy on the condition that they can't have any say over operations, with selling ARM being their only way of influencing them in any way?

Of course, owning a company where you have no control at all isn't great, but in this case it might be worth it if Apple trusts ARM to keep doing well without Apple's help, and if it would prevent someone like NVIDIA from shutting Apple out.

And would there be any antitrust issues if Apple bought a 500 year license to freely (or at a pre-set calculation of pricing) use any and all current and future ARM designs?

It would be like a gas station refusing to sell gas. Utter suicide for ARM.
It's not just patents.

ARM has previously shut down open source software emulation of parts of the ARM ISA.

For a while QEMU could not implement ARMv7 I think, until changed their mind and started permitting it. There was an open source processor design on opencores.org that got pulled too.

The reasoning was something like "to implement these instructions you must have read the ARM documentation, which is only available under a license which prohibits certain things in exchange for reading it".

To my knowledge, "Apple Silicon" or what ever you want to call it is an ARM64 ISA, with Apple extensions

Third parties have been able to get IOS running in an emulator for security research and Android(!) has even been ported to the IOS devices that contain the "Checkrain" Boot ROM exploit (though with things like GPU, Wi-Fi etc in varying states of completion)

ISA is just about the instruction set that the silicon is designed to interpret. Aww compatibility does not imply any shared heritage in the silicon between apple and arm LTD cores. Just like AMD and Intel share no silicon design even though they are both sell processors implementing the AMD64 ISA.

(The silicon implementation of an isa is referred to as the microarchitecture btw)

Right, but the ARM ISA is covered by various patents, so you have to get a license for it. ARM has aggressively shut down any open efforts to make ARM compatible cores (without licenses).
Apple has an "ARM Architecture License". They have a custom implementation that's very different from the reference design (which is why they're miles ahead of other ARM CPU makers). I'm sure the license has contractual obligations that protects Apple. In short, Apple is in no danger, most likely.
Let’s ignore that founding ARM investor Apple has a perpetual license and their own CPU designs.

Do you really think NVidia is going to substantially overpay for ARM to then nuke it’s remaining value by going to war on it’s licensees?

You're swallowing the poison pill with pride. The reason why no ARM customer actually wants to buy ARM is because there is almost no way to make money off of it. If you cancel licenses ARM is going to implode and $32 billion worth of value locked up in ARM will vanish into thin air. If you want to extract value out of ARM you'll have to play nice and that is definitively not worth $32 billion either.
(Re: RISC-V) I hope so!!!
I hear a lot of people talking about RISC-V but I wonder why people don't think companies would move to other open ISAs like Power or MIPS.
"MIPS Open" came after the RISC-V announcement, and is still currently somewhat of a joke. Half the links on the MIPS Open site are dead.

I think one of the major points for RISC-V was to avoid the possibility of patent encumbrance of the ISA so that it can be freely used for educational purposes. My computer architecture courses 5-6 years ago used MIPS I heavily. MIPS was not open at the time, but any patents for the MIPS-I ISA had long since expired.

POWER is actually open, but it is tremendously more complicated. RISC-V by comparison feels like it borrows heavily from the early MIPS ISAs, just with a relaxation of the fixed sized instructions and no architectural delay slots and a commitment to an extensible ISA (MIPS had the coprocessor interface, but I digress).

The following is my own experience - while obviously high performance CPU cores are the product of intelligent multi-person teams and many resources, I believe RISC-V is simple enough that a college student or two could implement a compliant RV32I core in an HDL course if they knew anything about computer architecture. It wouldn't be a peak performance design by any measure (if it was they should be hired by a CPU company), but I think that's actually a point of RISC-V as an educational platform AND a platform for production CPU cores.

As a teaching tool RISC-V is clearly great, as it is for companies that want to add custom instructions to their microcontrollers like NVidia or WD. But if I was looking to design a core to run user applications then to me it looks like everything is stacked in favor of Power. The complexity of the ISA is dwarfed by the complexity of a performant superscalar architecture. And to be performant in RISC-V you'd probably be needing extensive instruction fusion and variable length instructions anyways further equalizing things. And you really need the B extension which hasn't been standardized yet. Plus binary compatibility is a big concern on application cores and ISA extensions get in the way of that.
> still currently somewhat of a joke. Half the links on the MIPS Open site are dead.

unsurprising given that Wave Computing shut down the project almost a year ago and subsequently declared bankruptcy.

Because they aren’t as open in reality as they sounded when announced.

The advantages of those two (and of ARM) is that there actual Implementations with decades of development behind them, Yes, some technical debt but also many painfully-learned good decisions.

RISC V, which I’m really excited about, you can think of as a PRD (customer’s perspective product requirements document). That’s what an ISA is. Each team builds to meet that using their own iomplemetation, none of which is more than a few years old yet. But the teams incorporate decades of learning , and have largely a blank sheet to start with. I think it will be great....but isn’t yet.

The MIPS Open ISA project was actually shut down after only a year. [^1]

POWER has technically only been "open" for a little under a year. OpenPOWER was always a thing, but this used to mean that companies could participate in the ISA development process and then pay license fees to build a chip. This changed last year when POWER went royalty-free (like RISC-V).

The real defniition of "open" is can you answer the following questions in the negative:

  - Do I need to pay someone for a license to implement the ISA?
  - Is the ISA patent-unencumbered?
RISC-V was the only game in town for a long time and thereby attracted large companies (including NVIDIA) and startups that were interested in building their own microprocessors, but didn't want to pay license fees or get sued into oblivion.

[^1]: https://www.hackster.io/news/wave-computing-closes-its-mips-...

I think you are confused of what RISC-V is and how these things work.

RISK-V is just ISA.

You will have open source RISK-V microarchitectures for processes that are few generations old. You can use the same design long time when the performance is not so important.

You will not get open source optimized high performance microarchitectures for the latest process and large volumes. These cost $100s of millions to design and the work is repeated every few years for a new process. Every design is closely optimized for the latest fab technology. They have patents.

Intel, AMD, Nvidia, ARM, all have to design new microarchitectures every few years. It's not just doing some VHDL design. It involves research, pathfinding, large scale design, simulation, verification and working closely with fab business. The software alone costs millions and part of it is redesigned. Timelines are tight or the design becomes outdated over time.

"Building" for the latest process and large volumes is another story, but as far as I can see, large scale logic design is something not _that_ far away from software. Large scale, open source, and performant software designs exist in the wild. (see Linux, llvm, ...)

Why wouldn't we get a logic netlist which could perform reasonably well when placed on silicon by people who know what they are doing? (Yeah, lots of handwave.) I'm asking this out of curiosity. Not an expert in the field by any means.

SonicBoom is an open source Risc V core that is clock for clock competitive with the best from ARM. There is still the issue of matching clock speed and tweaking for a given process.

Foundries actually have an interest in helping to optimize an open core for their process as a selling point since it can be reused by multiple customers.

Their own paper[0] shows it performing at similar levels to the A72, a four year old core. Those are obviously really impressive results for an open source core.

The best ARM core is Apple’s Lightning, which has an IPC rate about 4X that of the A72. [0]: https://carrv.github.io/2020/papers/CARRV2020_paper_15_Zhao....

Do you have a reference the Apples IPC?
I was using AnandTech’s SPEC2006 benchmarks.
RISC-V is in danger of imploding from its near-infinite flexibility.

It is driven largely by academics who lack a pragmatic drive in areas of time-to-market, and it is being explored by companies for profit motives only. NXP, NVIDIA, Western Digital see it as a way to cut costs due to Arm license fees.

RISC-V smells like Transmeta. I lived through that hype machine.

Transmeta as in a company that was strong armed by Intel in a court case Transmeta could have won if they hadn't run out of cash and time (and they ran out of those things because of that court case)?
Transmeta was never a viable solution: it was a pet project billed as a disrupter on the basis of it being a dynamic architecture. Do I need to explain the mountain of issues with their primary objectives or do you want to google that?
I imagine Apple already negotiated a long-term (or in perpetuity) license for ARM so Nvidia won't be able to put the squeeze on them...
apple is going to be pissed.
This would be good. I'm waiting to see high performance PCs running on ARM CPU instead of x86.
Looking further into the future nvidia buying arm now means they could have their cpus and gpus in the next generation of consoles instead of amd
Except AMD has:

* better single threaded performance than ARM

* the same instruction set as developers' workstations

* the willingless to add Microsoft/Sony's hardware IP to the console chips

1) There are many domains that don’t care about single threaded CPU performance or CPU performance at all. I would say gaming is becoming one such domain

2) This might soon change if Microsoft and Apple succeed in their quest to push ARM machines to the consumer market. I can imagine than an ARM macbook is all that's needed to start an avalanche in adoption

3) Willingness can change easily :)

Other than the same instruction sets the other two factors can change in the 5 to 10 years that a console generation lasts.
Why is ARM selling itself? May be I am missing something here but if Qualcomm can generate a bulk of their revenues through licensing fees, shouldn't ARM be bringing in more? It feels like going public would be a profitable route for their investors. Wonder why they are opting for that.
ARM is owned by SoftBank, who has had a very bad year and a half. I guess they just want cash, not to take it public.
The irony of WeWork causing a great product like ARM to need to be sold off...
Irony redux: even if WeWork wasn't a trainwreck, this pandemic would have killed it.
It wouldn't have been great in the very short term. But there's an argument that mid-long term, the pandemic will be good for WeWork.
And what would that argument be? That the increase in unemployment leads to more new businesses being founded that can somehow miraculously afford luxury(-priced) offices?
That a lot of businesses are deciding they don't need as much full-time office space as it seemed like last year. Long term leases are expensive. On-demand, short-term leases are looking more attractive.
(comment deleted)
(comment deleted)
Softbank who is completely in the hole due to their continuously bad investments is the seller :)
ARM is wholly owned by an investor called SoftBank, who is desperate for cash right now, because they messed up with WeWork.
ARM isn’t selling itself. SoftBank bought ARM several years ago. I would imagine all the cash they hemorrhaged on wework has finally forced their hand.
Wireless technology is a patent minefield. CPUs generally aren't because the techniques for internals are old, and internal design, manufacturing, and especially validation (where intel fell flat lately) methodologies dominate the costs and quality.

For much of ARM's recent volume use in MCUs and slightly larger embedded devices, there is credible threat from first party usage of RISC-V (see WD, Nvidia).

Access to the ISA itself can be of high value.. see x86 and s390x for prime examples. Although I don't really see how ARM could pull that off outside being an acquisition like this, and making the licensing process onerous enough that people move to buying chips from nvidia instead of doing their own designs. In such a scenario, RISC-V can become a credible threat to phones too, and the server thing pundits keep pushing for the past 15 years never happens.

So there is a lot of value here, but it's pretty hard to grow as a pure licensing play as ARM has been since there are many risks and opportunities for price compression.

I wonder what this means for Tegra and/or mobile GPUs.
(comment deleted)
I bet Apple's really banging their heads against a wall right about now.
Can anyone explain what is going on between Apple and Nvidia? Seems Apple will not add Nvidia hardware to its machines no matter what. What’s the back story to that?

And where this is related is I wonder if Apple will have to relent (assuming the purchase goes through) and do business with Nvidia since it licenses some technology from ARM. Or, have I got it wrong here? Does Apple not rely on ARM?

(comment deleted)
NVIDIA really ratfucked them with the GPUs in MacBook Pros maybe a decade ago. They started failing like crazy, NVIDIA said they fixed it, they hadn’t, and then when things went to litigation NVIDIA blamed Apple for the GPU failures. Apple ran a giant recall and swore off using NVIDIA ever again.
A little tidbit, and I of course don’t have any proof to back this up but: for the eligible Mac models to qualify for a free mlb replacement from apple, the computers HAD to fail a specific diagnostic (called VST - video systems test). The repair couldn’t be classified as “covered” unless there was a record of this test failing. I’m also certain that privately nvidia agreed to cover the cost (or some of the cost) of these repairs, but stipulated they would only cover the computers that failed the test. Most computers did fail the test, but I definitely saw some machines that absolutely were experiencing gpu issues but wouldn’t have been eligible (most of the time Apple did the right thing and paid for it out of pocket)

Sorry for rambling! Just thought it was an interesting tidbit

Google "macbook faulty nvidia". e.g. [1] references to "arrogance and bluster" from Nvidia.

My personal experience: I had a 2011 Macbook Pro with an nvidia card. It started to fail randomly. Apple identified that certain nvidia GPUs were failing and created a test the "Geniuses" would run. My Macbook always passed, even though it kept throwing noise on the screen anywhere other than the Apple store. Eventually it finally failed their test: Four days after the (extended) warranty period. They refused to replace it. Bitterly, the best option for me was to pay the $800 for a new board.

[1] https://appleinsider.com/articles/19/01/18/apples-management...

Wow. I have had Apple do me right for free repairs out of warranty (and past AppleCare expiration) but several things applied: It was in California where retail culture is less combative about helping the customer; 2) the laptop did have AppleCare; 3) It was at the Palo Alto store which tends to have very well educated employees, with cross flow between the employee pool and corporate; 4) the laptop was purchased at an Apple store, which inexplicably gives them more leeway (well, possibly due to their certainty it was always in Apple’s hands prior to direct sale to the customer).
For what it's worth, I had almost exactly the same experience with the same 2011 MacBook as 'lowbloodsugar'. From my point of view, it was clearly starting to fail while still in AppleCare warranty (snow on the screen when doing anything graphics intensive), but somehow it passed their 'tests'.

Shortly after the warranty ended, the Nvidia card failed for good. On the bright side, it was a dual-video system, and I was able to keep it running a few years longer through complicated booting rituals that convinced it to boot with the lower power Intel graphics (although I lost the ability to recover from sleep).

I'm not sure why we have such different experiences with Apple customer service. I was also in the Bay Area, but not Palo Alto. It's possible that with a more aggressive approach I could have gotten them to fix it. Instead, I accepted their verdict, and now spread the word through forums like this that their warranties are not to be trusted.

Interesting, I have nearly the exact same situation except replace Nvidia with AMD. "Radeongate".

Also have miraculously saved the machine with a booting ritual/hack that bypasses the discrete gpu

Also 2011 MacBook.

I'm worried that you are right, and that I had an AMD card as well. I'm unsure, as the focus of my disappointment is Apple, rather than the manufacturer of the card.
That’s a super, super shitty experience. I made a comment somewhere else in this thread outlining my theory that apple were so strict with the test because part of the financial arrangement with nvidia stipulated the machine had to fail a diagnosistic they provided or contributed to.

Still, as someone who used to work as a “genius”, it would have been easy to cover your mlb repair under warranty for a million reasons, so the technician failed you there, and then to not do anything for you 4 days outside of eligibility especially based on your experience... that’s fucked. Not all “genius” bar people are like that. If something similar happens in future, call AppleCare and ask to speak to “tier 2”. They sometimes have leeway to issue coverage in extenuating circumstances.

I hope Huawei beat them to it.
This makes no sense. nVidia is already transitioning away from ARM and shipping RISC-V controllers in their GPUs. They already gave up on Denver, their high perf ARM server chip. Why would they buy ARM?
The would buy ARM because they believe it can result in long term profits.
Pure uneducated guesses from me: maybe the reason for giving up on Denver would no longer be a reason if they owed ARM. Or maybe they see a future use of ARM's tech that will either benefit them to own, or that scares them that ARM competition (or competition from other companies working with ARM) in a current or future area. Or just that they believe ARM is a solid business that will make money whoever owns it.

But I don't have a clue if it's a good idea or not.

Controlling the ARM holdings the company is very different value prop than a single arm chip .

When they made those decisions ARM wasn’t available for buying, SoftBank was ok top of the world didn’t have to sell anything . A year back do you think SoftBank would consider a proposal so close to their purchase price ?

Because the investment bankers desperate for fees said they will to the press to try to gin up other buyers?
They're using RISC-V for their micro controllers. I don't think we have any reason to believe they won't stick with ARM for application cores. And they've tended to ping-pong between Denver cores and standard ARM cores so I wouldn't entirely write Denver off.
Such a good use of their inflated stock price for an all-stock transaction. Go, NVIDIA!
SoftBank would want cash, they need to reduce their debt load , that’s pretty much they are considering this price in the first place
(comment deleted)
Fortunately there's a market for stocks.
There is no easy for market for 32b of any stock . You cannot just make sell order for that qty and expect that Order to be filled.

That stock market is also there for ARM stocks . There is no need to trade for nvidia shares at an undesirable price if there is no immediate cash in it.

Nvidia could potentially raise money from an FPO or similar instruments and leverage advantageous stock price they have, I.e. issue lesser newer stock than they would have at a lower price ,but only way SoftBank will consider a deal is if they get cash

Sure you can, someone will buy that $32B in nVidia stock for $20B. Which is still more than ARM is worth.
There actually is. For example - on the last annual Berkshire shareholders meeting Buffett was talking about how easy it was for them to unload billions worth of airlines, 100% of their holding.

On a typical single day $4B worth of NVIDA stock is traded.

PS. ARM is currently privately held (i.e. it is not on the stock market).
(comment deleted)