Howdy, Stranger!

It looks like you're new here. If you want to get involved, click one of these buttons!

RIP ARM

QuizzicalQuizzical Member LegendaryPosts: 24,845
edited September 2020 in Hardware
Nvidia today announced that they're going to buy ARM for $40 billion.

https://www.anandtech.com/show/16080/nvidia-to-acquire-arm-for-40-billion

The deal makes sense for Nvidia, and it makes sense for ARM.  The only problem is what happens to the rest of the tech industry.  If no regulators block the merger, then it will be hard to understand why they review mergers at all.

ARM's business is licensing IP.  They design CPU cores and GPUs, and they get them working on advanced process nodes.  If you want to use ARM's CPUs or GPUs in your own designs, all you have to do is license the IP.  They're very good at this, which is why ARM CPUs have come to dominate the industry.

Nvidia's business isn't licensing IP.  Rather, their business is in designing chips and getting them fabricated.  Enormous, expensive, and very high performance chips.  Mainly Nvidia does GPUs, but they're working on getting into other markets, too.

There's no intrinsic reason why Nvidia can't license IP.  They've tried to license their GPUs in the past, the way ARM, Imagination, and AMD do, but without any success.  The problem is that Nvidia has a thoroughly deserved reputation for being an awful business partner who will squeeze you for more money whenever they see an opportunity.  They've burned a whole lot of bridges this way.

Microsoft and Sony have each used an Nvidia GPU for a console generation exactly once.  And been burned by it, learned their lesson, and never licensed a GPU from Nvidia again.  They have no problem with licensing GPUs from ATI/AMD, and have each done so several times.

Nintendo learned from their competitors' mistakes, and when they went with Nvidia for the Switch, they didn't pay Nvidia for a custom chip.  Rather, they used a completely off the shelf Tegra X1 chip that Nvidia had already designed.  Microsoft and Sony have AMD customize all sorts of stuff for them in their Xbox and PlayStation chips.  Nvidia would have done the same if asked, but Nintendo knew better than to ask.

Nvidia has also burned bridges with Apple.  They used to use Nvidia GPUs sometimes and AMD GPUs sometimes in their Mac products.  Then Nvidia sold Apple a bunch of defective chips, and Apple has been exclusively AMD for discrete GPUs for several years.

It's not that AMD had better GPUs than Nvidia the last several years.  Much of the time, they had significantly worse GPUs.  But Microsoft, Sony, and Apple all knew that if they license something from AMD, then AMD will play nicely.  If they try to use an Nvidia GPU, Nvidia won't play nicely.  So the only choices are AMD and Intel, at least if Intel can eventually make a good GPU.

And that's to say nothing of the many cell phone vendors that Nvidia has burned bridges with.  Nvidia made promises about what their Tegra chips would do, cell phone vendors designed phones to use the Nvidia chips, and then the chips showed up burning massively more power than promised.  Nvidia announced dozens of design wins to get Tegra into cell phones, none of which ever made it to market.

So for much of the tech industry, Nvidia owning ARM is definitely a nightmare scenario.  That gives Nvidia the ability to cripple your products whenever they feel like it.  Nvidia doesn't build products in most of the markets where ARM cores are used, but they could readily change that.  If they control the CPUs that all of their competitors use, then Nvidia would have the ability to take over selected markets whenever they get around to it.

And the rest of the tech industry knows it.  If the acquisition goes through, expect a stampede away from ARM.  You know how Apple announced that they're moving Macs from x86 to ARM?  Not going to happen if ARM is owned by Nvidia.  You'd better believe that Apple is already trying to figure out what else they can move their CPUs to--including moving iOS devices away from ARM.

So why would Nvidia pay $40 billion to buy ARM and run it into the ground?  Don't think of it that way.  Think of it as paying $40 billion for the ability to massively disrupt all of your competitors in markets that you'd like to enter, at a time and in a manner of your choosing.

So what happens if the deal doesn't go through?  Well, that works out very badly for ARM, too.  You think you've sold the business to Nvidia, and the rest of the tech industry spends a year trying to figure out how to get away from using your products, and then you don't end up getting the money from Nvidia.  That doesn't necessarily bring you back to the status quo, as when all of your customers have just spent a year desperately trying to figure out how they can stop buying anything from you, some of them might have succeeded.  Plans put in place to replace ARM by something else might yet be executed.  That might not be an unrecoverable event for ARM, but sure doesn't seem likely to work out well for them.

So basically, if Nvidia buys ARM, then ARM is dead, at least beyond whatever proprietary uses Nvidia gets out of using ARM cores in products that they design themselves.  If Nvidia doesn't buy ARM, then ARM is still in serious trouble.  Because this has about as much chance of working out well for ARM's customers as the acquisition by AOL did for Netscape users.
harken33Gdemami[Deleted User]

Comments

  • botrytisbotrytis Member RarePosts: 3,363
    For NVidia it makes sense. Why? Because they will be able to manufacture super-computers with there GPU/CPU lay out which is where NVidia is making their money at.
    harken33


  • harken33harken33 Member UncommonPosts: 78
    Nice post Quizz.

    Since i generally have no inkling what major corporations are worth i was sure it must have been a $40 billion merger not an acquisition. Imagine my surprise to learn they are worth upwards of $320 billion, $100 or so billion more than Intel!

    Now i am at work and cant stop checking out tech companies value. What a rabbit hole you have sent me down :)
  • RelampagoRelampago Member UncommonPosts: 420
    For Nvidia it makes a ton of sense.  For ARM its licensees will be...wait for it...up in arms.
    botrytis
  • VrikaVrika Member LegendaryPosts: 7,577
    edited September 2020
    Quizzical said:
    ....

    So why would Nvidia pay $40 billion to buy ARM and run it into the ground?  Don't think of it that way.  Think of it as paying $40 billion for the ability to massively disrupt all of your competitors in markets that you'd like to enter, at a time and in a manner of your choosing.

    ...
    Now you're just blinded by your NVidia hate.

    NVidia is buying ARM because they want to have their own CPU tech in addition to GPU tech. Also NVidia seems to be better at making money, whereas ARM has been in a situation where everyone uses their CPU but ARM can't translate it into income. NVidia likely genuinely believes that they can do better with ARM (income-wise, even if it won't be as popular).
    botrytis
     
  • QuizzicalQuizzical Member LegendaryPosts: 24,845
    Vrika said:
    Quizzical said:
    ....

    So why would Nvidia pay $40 billion to buy ARM and run it into the ground?  Don't think of it that way.  Think of it as paying $40 billion for the ability to massively disrupt all of your competitors in markets that you'd like to enter, at a time and in a manner of your choosing.

    ...
    Now you're just blinded by your NVidia hate.

    NVidia is buying ARM because they want to have their own CPU tech in addition to GPU tech. Also NVidia seems to be better at making money, whereas ARM has been in a situation where everyone uses their CPU but ARM can't translate it into income. NVidia likely genuinely believes that they can do better with ARM (income-wise, even if it won't be as popular).
    Oh sure, Nvidia is going to use ARM cores.  They've been frustrated by their inability to get Intel or AMD to offer more bandwidth to connect their GPUs to the CPUs, and controlling the entire platform allows them to do whatever they want.

    But for ARM's current customers, needing to license your CPUs from Nvidia would be an unusually bad situation.  Expect a lot of ARM licensees to bail out as soon as they can (which could take years) if this goes through.
  • RidelynnRidelynn Member EpicPosts: 7,375
    edited September 2020
    Gotta say I couldn't see nVidia pulling off the acquisition and having it make any sort of sense. Now to see what they do with it. nVidia and their investors are really bullish about this one.
  • RidelynnRidelynn Member EpicPosts: 7,375
    edited September 2020
    Quizzical said:
    Vrika said:
    Quizzical said:
    ....

    So why would Nvidia pay $40 billion to buy ARM and run it into the ground?  Don't think of it that way.  Think of it as paying $40 billion for the ability to massively disrupt all of your competitors in markets that you'd like to enter, at a time and in a manner of your choosing.

    ...
    Now you're just blinded by your NVidia hate.

    NVidia is buying ARM because they want to have their own CPU tech in addition to GPU tech. Also NVidia seems to be better at making money, whereas ARM has been in a situation where everyone uses their CPU but ARM can't translate it into income. NVidia likely genuinely believes that they can do better with ARM (income-wise, even if it won't be as popular).
    Oh sure, Nvidia is going to use ARM cores.  They've been frustrated by their inability to get Intel or AMD to offer more bandwidth to connect their GPUs to the CPUs, and controlling the entire platform allows them to do whatever they want.
    nVidia has already been an ARM customer for quite some time. Tegra first came out in 2008, and unless anyone forgets, they power the Nintendo Switch, among other high profile accounts.

    To take nVidia at face value - this was a datacenter AI play. I think that's really the only way this makes any sense, and even then only if you are extremely bullish on what SOCs do for you there, as opposed to what nVidia already does with their GPU tech.

    I tend to agree with Quiz though - this is less about getting the tech, because nVidia already owned a license to the tech, nor is it about lowering your own licensing fees, as $40B makes for quite possibly the most expensive licensing fee in history... but more about being able to disrupt the competition if they start to muscle in too closely. You also can use it to force your proprietary tech down their throat: no more Mal or Imagination GPU cores, now you must use nVidia GPU cores with your ARM license, or pay for the privilege of doing without.

    Those last two paragraphs may sound like I'm contradicting myself, but I don't see them as exclusive to each other - rather I think they are very much intertwined.

    It also makes one heck of a poke in the eye to Intel, who could never quite get in on the mobile stack and wasn't willing to let nVidia come and play with x86... now in one stroke nVidia has not only caught up, but leapfrogged Intel in the largest CPU market in the world.

    The biggest question I have is how will this apply to current licensees? That's how Apple will really respond: their current CPU plans are ARM.. but if they diverge away and keep iterating on their own CPU design while keeping their own GPU tech, how far can they take that without renewing the ARM license?
  • QuizzicalQuizzical Member LegendaryPosts: 24,845
    Right now, there are two very different types of ARM licenses.  One is that you can license a completed ARM core and use it however you want, but not change the core itself.  Lots of companies do that.

    The other is an ARM architecture license.  That allows you to make up your own ARM core and use it.  You can change cache sizes, decode widths, number of ALUs per core, or whatever.  Apple has done that most famously.  Qualcomm has done it a lot, too, and Nvidia did for one generation.

    What you cannot change with an architecture license is the ISA, however.  That is, you cannot add new instructions, nor remove ones that you don't want.  All ARM v7 cores must have exactly the same instructions available.  They must all be able to run exactly the same code, even if some run it faster than others.  ARM v8 can modify what ARM v7 had, but ARM wanted to prevent the cores from becoming highly fragmented so that end users had no idea which code could run on which cores.  Basically, ARM didn't want the ARM ISA to end up like AVX-512.

    If Nvidia owns ARM, they own the architecture.  They can add absolutely any instructions that they want.  They can drop instructions that they don't want.  Any instructions that they think will help with bandwidth to a GPU can be added.  Stuff that they think is useless in a data center can be dropped.

    Right now, the ARM architecture is mostly designed with low power use in mind, not data center servers that use several thousand watts.  Nvidia can change that if they want.  ARM sure wasn't going to.

    That doesn't necessarily mean leaving cell phones and embedded out in the cold.  Nvidia could fork the architecture.  They could make an Nvidia-only version of the architecture with data center stuff that no one else is allowed to use, while the main ARM architecture remains focused on more traditional uses of the cores.  Owning ARM lets Nvidia do a lot of things that right now, they can't.

    So even without wanting to mess with competitors, Nvidia does have a clear case of why owning ARM benefits them.  That they can use it to mess with competitors is just a bonus.

    I think that the ultra low power uses of ARM cores are going to continue using older ARM cores for some years, then ultimately go away if Moore's law lasts long enough.  Right now, RISC-V seems like the most promising contender to take that market segment.

    While ARM will be harder to replace is in the higher power uses of their cores.  Right now, the only architectures that can compete with Apple's custom ARM cores are x86 and Power, and neither of those are for sale.  Neither are appropriate to iPhones, either, and Apple would really like to use the same architecture in everything from an Apple Watch to a Mac Pro.

    It will be interesting to see what Apple does.  Unless they can cut some agreement with Nvidia that allows them to do absolutely whatever they want with ARM cores forever, I suspect that the Mac move to ARM is dead, or at least short-lived.  Apple is not going to be willing to rely on the good graces of Nvidia in order to make future products forever.

    The question is where Apple goes instead.  Apple has the ability to create their own architecture from scratch if so inclined, and doesn't place a high value on interoperability with hardware from other vendors.  They could also try to buy Imagination, which owns MIPS.

    It will also be interesting to see what happens to ARM servers.  Most ARM CPUs aren't intended for data centers, but a Marvell ThunderX2 or Amazon Graviton2 absolutely is.  Those are direct competitors to Nvidia's stated, intended use of ARM.  And they really don't have any alternatives to ARM.  Trying to license whatever Nvidia will let you have in order to make future products that compete with Nvidia's is not a serious option.

    The final interesting market to watch is cell phones, or perhaps rather, Android phones, since Apple can go a different direction.  That's not a market where Nvidia plays right now, but they have tried to get into it in the past.  Once they own ARM, they could easily try again.  Moving the entire Android ecosystem to an entirely different architecture doesn't seem easy, but making all Android hardware completely dependent on what Nvidia feels like charging isn't a viable option, either.

    There are a lot of companies who have a lot to lose from Nvidia buying ARM.  Just listed above (sometimes implicitly) is Apple, Amazon, Google, Samsung, Qualcomm, Marvell, and some others.  Expect to see a lot of lobbying over whether regulators should approve the sale.
Sign In or Register to comment.