A NVidia purchase of ARM would also create a lot of conflicts of interest.
ARM is more of a household name for their use in mobile phones but that is just the tip of the iceberg.
I think you underestimate how many ARM chips you have in a single car or delivery truck.
Add to that:
- farming machinery
- construction machinery
- factory line automation
- elevators, escalators
- EV chargers
- fridges, washing machines, ovens
- medical devices such as drug-infusion pumps, ventilators, surgical machinery, etc.
- Auxiliary modules in aeroplanes and shipping containers.
- Infrastructure for Road, Rail, Power Grid with ARM processors running headless embedded systems
- Anything the size of a pebble with bluetooth connectivity uses nordic's nRF chip (which is yet again an ARM chip)
ARM processors are hiding in plain sight in the world all around you.
I understand the point you make, from a business standpoint, on how Tiktok might scale. The thing that is bizarre for me and I agree with the parent's sentiment, is how disconnected the valuation is from the real-world impact and objective _usefulness_ of ARM versus Tiktok.
Edit: bullet points
Also how easily the technology behind Tiktok can be duplicated compared to ARM.
Anyway, it's probably all about the network effect and brand value.
Tiktok can deliver arbitrary content to hundreds of millions of people’s eyeballs around the planet.
Consider that in 2017, 21 billion ARM chips were manufactured (doubled in 4 years, from 10 billion in 2013), and that ARM's licensing fees are over 2% of chip cost for current high-end designs (and they're talking about raising that even more). They have 95%+ of the mobile phone market, are making inroads in the server market, and will soon be in every Apple laptop, which I expect will grow the market for ARM laptops even outside Apple. It wouldn't be a stretch to find them in common desktop computers after that. They're in smart TVs, washing machines, robot vacuum cleaners, and all other kinds of smart (and non-smart) appliances. Even SSDs have their own embedded ARM chip. And that's just home/consumer stuff; haven't even scratched industrial/commercial applications, of which there are a ton.
I could easily see yearly production at over 100 billion chips before 2030, probably even before 2025. While I'd love to see something like RISC-V take off commercially, I don't think that's realistic.
Meanwhile, social media users are incredibly fickle; platforms are subject to fad and fashion. Certainly Facebook and Instagram are still huge behemoths, but their growth is nothing like it once was, with people -- especially younger people, trying to distinguish themselves from their older, boring relatives -- flocking to TikTok. I fully expect TikTok will be in Instagram's boat in under 10 years, with some other platform taking its place.
ARM seems like an amazingly great short-, medium-, and long-term bet, while TikTok feels like a nice short- (maybe medium-, if they're lucky) term money-maker, and even that feels like a big maybe: I have no idea what their ad revenue per user looks like, but it's probably not great since their audience skews younger. Teenagers and college kids don't have much in the way of discretionary income. Then again, TikTok doesn't pay their creators like e.g. YouTube does, so they get to keep all that ad revenue.
Now, certainly TikTok might not be sustainable and might disappear off the face of the earth tomorrow. Or it might become a juggernaut that overtakes Facebook.
> The actual rates an advertiser pays varies, usually between $0.10 to $0.30 per view, but averages out at $0.18 per view.
https://influencermarketinghub.com/how-much-do-youtubers-mak...
This also is only people that watch the whole ad.
PS I applaud RISC-V but it won't take over the market for a long time, and it wouldn't drive ARM out completely, I'm sure. Intel's had many competitors and they're doing just fine (even despite screwing up repeatedly with their processes!)
Look at all the failed attempts to move away from x86(/64). Even intel tried it with Itanium and failed, HP has to pay them to keep making it so they can fulfull their server contracts. I'm sure ARM has a similar hold on the mobile market.
With AWS offering ARM systems, all the Chromebooks, Apple, the complete loss of the phone market, Intel’s staying power is about to be tested to the extreme.
The only actual major change here is AWS offering Graviton, which actually hints at their real cash cow: datacenter SKUs with absurd markup. Something like 80% of their profit margins are here. More accurately, the change is that there are now viable silicon competitors to Intel in the performance department. So it's now clear that ultra-integrated hyperscalers who can actually afford tape out costs (7nm CPUs are not cheap to produce in volume) have an option to vertically integrate with e.g. Neoverse. Smaller players will not do this still, because alternative options like Rome will be adequate. But the only reason any of them are changing anything is cost savings, because now there are actual viable competitors to Intel when there were zero of them for like, 15 years. Producing cutting edge silicon products isn't easy, but it's very profitable, it turns out.
To be clear, Intel isn't charging $10,000 for a Xeon Platinum because it costs $9500 to make and they make $500 in profit. (Likewise, AMD doesn't produce competitors at 1/5th the price because they made a revolutionary, scientific breakthrough in processor design.) They're charging what you'll pay, not what it takes to produce. Seeing as they currently still have a complete stranglehold on the datacenter industry and make more in a quarter than most of their competitors do in several years, I suspect they've got much more "staying power" than the watercooler chat on this website would lead you to believe.
The Softbank guy invested a ton of money on WeWork. Tried to sell WeWork for 60 billion, but before that happen WeWork valuation dropped out to 2-5 billion (huge loss). That was in 2019. Afterwards, Softbank invested another 10 billions to try to save it. WeWork owns and also pays rent or hundreds of office buildings in the most expensive zones of all the major capitals in the world. 2020 COVID now means these super expensive offices are now empty, since WeWork customers pay a premium to be able to cancel their leases in <1 week. So essentially, WeWork is broke, worth 0, and Softbank has lost dozens of billions on it.
On top, Softbank owns a huge chunk of Uber, which is also worth close to zero now that people are not travelling due to COVID...
So... yeah... Softbank is selling ARM because they must. They are super broke, and investors are going to pull the money that remains out. Selling ARM and giving investors a tiny benefit so that they keep their money is better than them taking a huge loss this year.
MIPS _should_ be dead, half the manufacturers of the chip have stopped. But, Imagination Technology still sell a considerable number to Apple every year.
[1]https://qz.com/741078/a-company-that-doesnt-really-make-chip...
Here's an estimate of youtube (not exactly tiktok, but not exactly not), revenue at 10 to 30 cents per view.
[2]https://influencermarketinghub.com/how-much-do-youtubers-mak...
I loved BeOS, but there was an even more fundamental problem that put a limit on its days: A failure to anticipate coming need for home computers to become more secure. At the same time that Microsoft and Apple were both working frantically to ditch their old single-user desktop operating systems and replace them with, in effect, spruced up versions of existing server/workstation operating systems, Be was trying to launch a brand new OS on the dying model. Had they survived even a couple years longer, they would have had to reckon with that, and they simply didn't have the resources to navigate such a fundamental transition.
Let's be real here: TikTok is a big social network, yes, but ARM owns most of the embedded market. Every smartphone your average consumer buys runs an ARM chip. And hardware is harder to replace than software.
This number is not unbounded. It will converge to some asymptotic limit.
> How much profit will Tiktok make on each of those views?
Around a tenth of a cent per view.
These huge valuations are purely because people don't do the math and don't know how the market works.
That's how this year's myspace, which people will have trouble remembering 5 years from now, can get a higher "valuation" than a large semiconductor company with a 30 year track record.
Investors and the financial sector are proving time and time again that they're unable to learn from their mistakes, through no "fault" of their own, because apparently it's human nature to just be horribly bad at this.
It amazes me that people think investors somehow learned anything from the dot-com bubble, given they've been repeating all of their other major mistakes every odd year or so.
Edit: https://www.businessinsider.com/microsoft-tiktok-donald-trum...
This is some very fresh news.
I think, it is because people are now so used to Apple and Amazon's trillion valuation, with Apple closing in to 2 Trillion, people think $32B is low or ( relatively ) cheap.
Reality is ARM was quite over valued when it was purchased by Softbank.
This might also be part of SoftBank’s fire sale which bought ARM for $32B just few years ago (2018?)
I always just assumed that interfacing proprietary IP with the GPL is a tricky legal business. One slip, and all your IP becomes open source.
Do you have a source explaining what licensing changes they would have to make and what impact would that have for Linux and Nvidia ? I'd like to read that.
Hence AMDs push for Mantle then Vulkan. The console like API is the carrot to get people to use an API that has an verification layer so that third parties can easily say "wow what a broken game" rather "wow this new game runs on Nvidia and not AMD, what broken AMD drivers".
Nvidia open sourcing their drivers completely destroys a large chunk of their competitive advantage and is so intertwined with all the IP of the games they have hacks for that I'd be surprised if they ever would want to open source them, or even could if they wanted to.
More docs would be nice though.
It was:
- all kinds of problems wrt. the integration of the driver in the Linux Eco system, including the properitary driver having quality issues for anything but headless CUDA.
- nvidea getting in the way of the implementation of an open source alternative to their driver
And they don't see the benefit to create new drivers.
I agree they should help out more.
We could care less about the hacks and cludges baked into the proprietary Nvidia drivers and firmware focused on DirectX powered gaming. The current path Nvidia has chosen with signed firmware locks out open source developers from much of the low level operations of their GPUs.
AMD does the exact same thing and always has. When you see shaders come down the wire you can replace them with better-optimized or more performant versions. It's almost always fixing driver "bugs" in the game rather than actual game bugs. And the distinction is important.
I do agree with you, but that element is something everyone has to do to remain competitive in games. Developers will only optimize for one platform (because they're crunching), and 9 times out of 10 that's a RTX2080Ti.
While yes AMD did similar things when they could, it was way less prevalent (if only because they didn't have the staff necessary to pull it off to the same degree).
Edit: Here's an example of some of the stuff I'm talking about that goes beyond shaders. https://devblogs.microsoft.com/oldnewthing/20040305-00/?p=40...
This persistent bit of FUD really needs to die. Yes, you have to be careful, but at this point it's ridiculously well-known what is obviously correct and what is obviously incorrect when dealing with GPL. I'm sure there are some grey ares that haven't been worked out, but avoiding those is fairly simple.
Nvidia is already in a weird grey area, releasing binary blobs with an "open source" shim that adapts its interfaces to the GPL kernel. As much as the Linux kernel's pragmatic approach toward licensing helps make it easier on some hardware manufacturers, sometimes I wish they'd take a hard line and refuse to carve out exceptions for binary drivers, inasmuch as those can sometimes/always be considered derived works.
Way to contradict yourself.
> but avoiding those is fairly simple.
[Citation needed].
> As much as the Linux kernel's pragmatic approach toward licensing helps make it easier on some hardware manufacturers, sometimes I wish they'd take a hard line and refuse to carve out exceptions for binary drivers, inasmuch as those can sometimes/always be considered derived works.
Maybe this is what needs to happen to force companies to change their mindset, but where I work, lawyers tell us to (1) never contribute to any GPL'ed based code, (2) never distribute GPL'ed code to anybody (e.g. not even a docker container), etc.
Their argument is: a single slip could require us to publish all of your code, and make all of our IP open, and to make sure this doesn't happen, an army of lawyers and software engineers and managers would need to review every single code contribution that has something to do with the GPL. So the risks are very high, the cost of doing this right is very high as well, and the reward is... what exactly ? So in practice this means that we can't touch GPL'ed code with a 10 foot pole, it is not worth the hassle. If I were to ask my manager, it will tell me that it is not worth it. If they ask their manager, they will tell them the same. Etc.
BSD code ? No problem, we contribute to hundreds of BSD, MIT, Apache, ... licensed open source projects. Management tells us to just focus on those.
For some older card generations (e.g. GTX 600 series) it was competitive with the official driver. But in every hardware generation since then, the GPU requires signed firmware in order to run at any decent clock speed.
The necessary signed firmware is present inside the proprietary driver, but nouveau can't load it because it's against the ToS to redistribute it.
Most GPU features are available but run at 0.1x speed or slower because of this single reason. Nvidia could absolutely fix this "tomorrow" if they were motivated.
https://www.phoronix.com/scan.php?page=news_item&px=Nouveau-...
At this point I've gotten so bloody tired of the games people play with IP, that I'm arriving at the point I think I wouldn't even mind being part of the collateral damage of our industry being burned to the ground through the complete dissolution of any software delivered or related contract. If you sell me hardware, and play shenanigans to keep me from being able to use it to it's fullest capability, you're violating the intent of the rights of First Sale.
To be honest, I think every graphics card should have to be sold bundled with enough information for a layperson (or I'll throw out a bone,a reasonably adept engineer) to write their own hardware driver/firmware. Without that requirement, this industry will never change.
The point is it's not about open sourcing your properitary driver but about not getting in the way of an alternative open source driver, maybe even letting it a bit of an have even if just unofficially.
I thing if I where nvidea I might go in the direction of having a fully it at least partially open source driver for graphic stuff and a not so open source driver for headless CUDA (potentially running alongside a Intel integrated graphics based head/GUI).
Through I don't know what they plan wrt. ARM desktops/servers so this might conflict with their strategies there.
The only tricky things involve blatantly betraying the spirit of the agreement while trying to pretend to follow the letter and hoping a judge supports your interesting reading of the law.
Even so there is no provision in law wherein someone can sue you and magically come into possession of your IP.
It would literally require magical thinking.
Take a look at a recent snapshot of changesets and lines of code to the Linux kernel contributed by various employers: https://lwn.net/Articles/816162/
Arm themselves is listed at 1.8% by changesets; but Linaro is a software development shop funded by Arm and other Arm licensees to work on Arm support in various free software, and they contributed 4% of changesets and 8.8% by lines of code. And Code Aurora Forum is an effort to help various hardware vendors, many of whom are Arm vendors, get drivers upstreamed, and they contributed 1.8% by changesets and 10.1% by lines changed. A number of other top companies listed are also Arm licensees, though their support may be for random drivers or other CPU architectures as well.
However, Arm and companies in the Arm ecosystem do make up a fairly large amount of the code contributed to Linux, even if much of it is just drivers for random hardware.
And Arm and Linaro developers also contribute to GCC, LLVM, Rust, and more.