News Arm Pioneer: Nvidia's Grace CPU Is Proof That It Will 'Compete Unfairly'

[-Fran-]

Bro, Intel DOES license x86 out to anyone that wants to use it. That's why AMD and VIA processors exist in the first place. You DO NOT want to be a monopoly in the United States, the US government will take your company away from you and smash it into little pieces right in front of your eyes. Intel has taken great pains over the years to keep AMD alive so that it doesn't look like a monopoly (Intel is one of AMD's largest stock holders).

Er... No... You're awfully wrong, so please check how the X86 ISA is divided between VIA, AMD and Intel.

This isn't comparable to Intel. Intel is the actual originator of x86 and they control the entire manufacturing cycle from R&D all the way to retail products. When you control the entire process, of course you don't want to license it out, you'd just be creating competition for yourself.

I mean, clearly... IBM forced Intel to licence X86 to AMD so they didn't get strong-armed by them. IBM is IBM for a reason. How ironic we have to thank IBM that AMD can make X86 CPUs.

Nvidia with ARM is not in the same position. The reason ARM is everywhere is because companies can customize it to their needs. NVidia can't bring the thousands of customized variations in house now and trying to consolidate them into a manageable number of variations is obviously not going to work for the 1800 existing ARM licensees. NVidia is also fabless. It would make zero sense for Nvidia to try and design ARM CPU's for Apple and Samsung and everyone else and then negotiate with TSMC to get them all produced. Nvidia is not buying ARM to go through all that.

The money is in the licensing. Let Apple and everyone else spend the money to develop their own CPU's and battle for FAB allotments, while NVidia sits back and collects checks. Apple can certainly afford to pay more for their license, and it will likely take a pretty large increase in price for them to decide to abandon ship and to start development all over on a different ISA.

Here are ARM's quarterly results

• Arm's quarterly net sales worldwide 2017-2020 | Statista

Income was about $1.9 billion in 2020, so I don't know what your link that says $1.5 billion is covering. There is no 4 quarter stretch going back to the beginning of 2017 where income was as low as $1.5 billion.

The Dollars And Sense Of Nvidia Paying A Fortune For Arm (nextplatform.com)

"Let that sink in while we consider how much money that is for a business that generated just shy of $2 billion a year in licensing fees and other revenues in SoftBank’s fiscal 2019 and 2020 years ending in March and had an income of $1.27 billion in fiscal 2019 and a loss of $400 million in fiscal 2020. The Arm division of SoftBank had a one-time gain of $1.67 billion in that fiscal 2019 year after setting up a joint venture in China and getting a big bag of cash. The point is, it is hard to say how profitable the Arm licensing business really is at this point. Right now, it really isn’t, "

Here is a link to Softbank's 2019 q4 financial report:

Consolidated financial report for the fiscal year ended March 31, 2020 (group.softbank)

Softbank themselves reported a negative operating income.

The Statista link says nothing of their profit. Also, the paragraph you quoted talks about Soft Bank and not ARM on its own. You need to look at ARM Holdings, not Soft Bank. They're different companies when filing yearly reports. ARM is still UK-based and doesn't file in Japan. Soft Bank can declare a loss, because it's off-set by ARM in the UK. Soft Bank is the parent Company, but not the actual Company we're discussing as they didn't move them out of the UK. Crazy how it works, but that's the way it is.

So no, ARM Holdings is not losing money. Not by a long shot and are profitable still.

If you want a proper break down of the whole relationship of Soft Bank, their stupid decisions and ARM, please read this:

https://asia.nikkei.com/Opinion/SoftBank-s-flip-of-Arm-Holdings-reflects-Son-s-short-termism

Regards,

 

[Soaptrail]

Let me make this very simple for everyone: if you have an opportunity to make your competitor’s life difficult, that’s what you do. Pure and simple.

It doesn’t matter if an argument is valid or not. If it costs you little or nothing... and if it forces your competitor to HAVE to answer and explain and expend resources... that’s what you do.

nVidia knows this is going to happen. Similarly, all of its competitions know their job. If they can stop this acquisition from happening, it doesn’t matter if it would’ve been benign or typical nVidia BS, they’ve managed to disrupt nVidia’s plans. And that’s a win.

It’s not personal. It’s just business.

That remind me of someone. I wish I could remember who...

 

[TJ Hooker]

Er... No... You're awfully wrong, so please check how the X86 ISA is divided between VIA, AMD and Intel.


I mean, clearly... IBM forced Intel to licence X86 to AMD so they didn't get strong-armed by them. IBM is IBM for a reason. How ironic we have to thank IBM that AMD can make X86 CPUs.

The Statista link says nothing of their profit. Also, the paragraph you quoted talks about Soft Bank and not ARM on its own. You need to look at ARM Holdings, not Soft Bank. They're different companies when filing yearly reports. ARM is still UK-based and doesn't file in Japan. Soft Bank can declare a loss, because it's off-set by ARM in the UK. Soft Bank is the parent Company, but not the actual Company we're discussing as they didn't move them out of the UK. Crazy how it works, but that's the way it is.

So no, ARM Holdings is not losing money. Not by a long shot and are profitable still.

If you want a proper break down of the whole relationship of Soft Bank, their stupid decisions and ARM, please read this:

https://asia.nikkei.com/Opinion/SoftBank-s-flip-of-Arm-Holdings-reflects-Son-s-short-termism

Regards,

ARM is a wholly owned subsidiary of Softbank, they don't file their own financial statements. Which is why there aren't any ARM-only financial releases after ~2016, when they were acquired.

The financial report spongiemaster linked did indeed contain values for ARM by itself, page 24. Their income was +134B yen in 2019, but -43B for 2020. The they also made 134B in 2018, but lost 31B in 2017, according to page 15 of Softbank's 2020 annual report. Note that these are fiscal years, not calendar years.

Edit: The values in that 2020 report I had linked don't line up with their financial reports, maybe it's based on calendar year or uses a different accounting method or something. The comparable values for 2017 and 2018 fiscal years can be found here (page 20). Income was +13B yen in 2017, -31B in 2018.

Edit2: Wrong units for values, supposed to be in billions not millions.

 

[-Fran-]

ARM is a wholly owned subsidiary of Softbank, they don't file their own financial statements. Which is why there aren't any ARM-only financial releases after ~2016, when they were acquired.

The financial report spongiemaster linked did indeed contain values for ARM by itself, page 24. Their income was +134M yen in 2019, but -43M for 2020. The they also made 134M in 2018, but lost 31M in 2017, according to page 15 of Softbank's 2019 annual report. Note that these are fiscal years, not calendar years.

Now that you mention it, I did have a look because something didn't make sense and you're correct. They have ARM Holdings as a dormant company in the UK so their filings are, well, empty. Like a note saying "done, just not here":

https://find-and-update.company-information.service.gov.uk/company/11299879/filing-history

Still, you've proven my point indirectly. Sure, it's not making a bazillion dollars/yen/shekles, but it's not in the red like implied.

Regards,

 

[spongiemaster]

Also, the paragraph you quoted talks about Soft Bank and not ARM on its own.

Are you trying to be funny? You read under $2 billion in revenue in 2019 and 2020 and you thought they were talking about all of Softbank? I'm not going to spend any time responding to anything else in your post, as I can't tell if you're trolling or what.

 

[TJ Hooker]

Now that you mention it, I did have a look because something didn't make sense and you're correct. They have ARM Holdings as a dormant company in the UK so their filings are, well, empty. Like a note saying "done, just not here":

https://find-and-update.company-information.service.gov.uk/company/11299879/filing-history

Still, you've proven my point indirectly. Sure, it's not making a bazillion dollars/yen/shekles, but it's not in the red like implied.

Regards,

Well no, it doesn't really support your point (and I've update the values for 2017/2018 in my previous post with a better source). They were marginally profitable in 2017, and lost money in 2018 and 2020. They did make money in 2019, but only because of a one time lump sum they got from a joint venture (as mentioned by the paragraph spongiemaster quoted earlier). So from their core licensing/royalties business they've lost money 3 of the last 4 years, with only minor profits the other year.

 

[-Fran-]

Well no, it doesn't really support your point (and I've update the values for 2017/2018 in my previous post with a better source). They were marginally profitable in 2017, and lost money in 2018 and 2020. They did make money in 2019, but only because of a one time lump sum they got from a joint venture (as mentioned by the paragraph spongiemaster quoted earlier). So from their core licensing/royalties business they've lost money 3 of the last 4 years, with only minor profits the other year.

Thanks for that.

Then why do I see a positive net income in FY17, loss in FY18? As I said, sure they're not making bazillions, but they're not unprofitable with their business model.

Also, this note: "Depreciation and amortization includes amortization expenses of ¥29,379 million for the previous fiscal year and ¥54,569 million for the fiscal year, which were related to intangible assets recognized in the purchase price allocation at the time of the acquisition of Arm." This quote is from the FY18 PDF you linked. That makes it sound like those "expenses" are due to the purchase of ARM (as in interest, conversion rates or something?) and not due to ARM's performance?

This is going on a hard tangent and we're not trying to examine the financial performance of ARM and while I didn't get the information correctly portrayed, I'm not that off the mark by saying that ARM is a profitable company still with the way they licence the ARM ISA (among the other things they do). We could see what is the most profitable part of it, but the point here is clear: nVidia doesn't want ARM because they're a the most lucrative company ever. All this discussion has made that even clearer, hasn't it?

Cheers!

 

[TJ Hooker]

Then why do I see a positive net income in FY17, loss in FY18?

Because they made money in 2017 but didn't in 2018, which is exactly what I said?

They were marginally profitable in 2017, and lost money in 2018

Also, this note: "Depreciation and amortization includes amortization expenses of ¥29,379 million for the previous fiscal year and ¥54,569 million for the fiscal year, which were related to intangible assets recognized in the purchase price allocation at the time of the acquisition of Arm." This quote is from the FY18 PDF you linked. That makes it sound like those "expenses" are due to the purchase of ARM (as in interest, conversion rates or something?) and not due to ARM's performance?

Not sure what that's referring to, but there's an equivalent note for 2019 and 2020. So whatever it is, it would seem to be an ongoing expense for ARM.

As I said, sure they're not making bazillions, but they're not unprofitable with their business model.
[...]
I'm not that off the mark by saying that ARM is a profitable company still with the way they licence the ARM ISA (among the other things they do)

How do you reconcile these statements with the fact they have not been profitable for the last three years? I'm treating 2019 as a loss given that the only posted a profit by selling off a majority stake in an ARM Chinese subsidiary to form the JV, which obviously cannot be considered part of their business model of licensing the ARM ISA.

 

[-Fran-]

Because they made money in 2017 but didn't in 2018, which is exactly what I said?

Oh, sorry; I missed that. In the last 4 fiscal years they're 2 and 2. Even if you pass FY19 as a "loss", they're still in the black.

How do you reconcile these statements with the fact they have not been profitable for the last three years? I'm treating 2019 as a loss given that the only posted a profit by selling off a majority stake in an ARM Chinese subsidiary to form the JV, which obviously cannot be considered part of their business model of licensing the ARM ISA.

Well, for 2 "soft" facts:
1- They're an old Company from the UK that has been making money for several years while not making for some others that has stayed afloat since the start in the 90's. Hence, whatever ARM has been doing is not that off the mark (edit: as in, they've managed to stay afloat).
2.- Their market cap in terms of IP licencing is so big that it's nearly impossible their operational income cannot make them stay in the blue/black* if they want to. They need to invest in R&D for sure, and I'm guessing most of their expenses are around there, plus when they need to expand.

And again, just to repeat it: this makes the point of nVidia wanting ARM not for their business model but the licencing even more evident, no?

Regards,

 

[Chung Leong]

One has to wonder how a company can have such a ubiquitous and dominant market share, about 95% of the cellphone market, and still be losing money. Apple is printing money and soon basically their entire product portfolio will be based off of technology from a company losing money.

From what I've heard, basically none of the Android phone makers makes any money. Except for the hyper-vertically integrated Samsung. Even then the profit margin is something like 50 cents per handset. There just isn't much money you can squeeze out of this market. Given the near complete market penetration, the growth prospect for such a strategy is definitely south of zero.

 

[InvalidError]

From what I've heard, basically none of the Android phone makers makes any money. Except for the hyper-vertically integrated Samsung.

From various tear-downs and analysis, the BoM+manufacturing cost of $1000+ phones is only around $275, plenty of headroom for profit there. On lower-end phones, manufacturing costs are closer to $150.

While Samsung may be technically capable of making nearly everything that goes into its devices, many of its phones use Qualcomm SoCs because Samsung isn't able to match Snapdragon's performance and power efficiency. For its lower-end devices, Samsung also uses cheap Mediatek SoCs instead of its own Exynos chips.

 

[Chung Leong]

From various tear-downs and analysis, the BoM+manufacturing cost of $1000+ phones is only around $275, plenty of headroom for profit there. On lower-end phones, manufacturing costs are closer to $150.

I would be surprised if the carriers pay anything above 50% retail price. In Q4 2020, Xiaomi's smartphone revenue was RMB42.6 billion, with 42.3 million units shipped. A thousand RMB is roughly $150.

 

[renz496]

Back in the 90's, the scuttlebutt was that RISC CPUs would "soon" be displacing CISC x86. Didn't happen then, when "x86" was actually something real. There are no "x86" CPUs today--neither Intel nor AMD makes CPUs on the original x86 ISA. Current "x86" CPUs today are RISC-CISC hybrids that carry a bit of the x86 instruction set for backwards compatibility purposes only, but which otherwise bear no resemblance to "x86" of the 80s/90s. The common misunderstanding is that current "x86" CPUs by AMD and Intel are static--sort of frozen and unchanging--but the reality is the exact opposite.

I think the UK would be well advised to nix the nVidia deal--over the last 25 years the company has shown a marked proclivity for cheating in various commercial and competitive memes and seems to have no qualms in using unfair and dishonest trade practices of all kinds. Microsoft found nVidia so disagreeable to work with that it dropped nVidia after the original xBox contract--which Microsoft had awarded nVidia. After the first xBox, Microsoft awarded the xBox contracts to ATi, now AMD, and left nVidia in its rear-view permanently. After 3dfx pioneered the first actually playable 3D GPU solutions in the late 90s, I watched nVidia level cheat after cheat against 3dfx in order to try and compete--as always, the cheating involved benchmarks almost exclusivity. Doesn't seem to have changed much, imo. The stories I could tell...!

nah MS actually work just fine with nvidia. the issue with xbox was the console was selling below expectation and then MS ask for nvidia to "lower" their initial agreed price so MS can cope with the lost which what nvidia does not agree with. if they really see nvidia as one they did not want to work with then Zune nor Surface for ARM will be using nvidia tegra.

 

[renz496]

When Nvidia bought Mellanox, a networking company, what was the aim behind that?
Is that why Geforce Now exists?

one of the main reason is not to be outdone by AMD and intel in HPC market.

 

[ClapTrapper]

A lot of visionaries believe that ARM is finally the one who will displace x86.

Nvidia is the biggest threat to that new future. What good is breaking from x86 when you replace it with the likes of Nvidia? Meet the new boss. Same as the old boss.

Earworm Alert!
Well,at least it is a great song!
;>

 

[Gomez Addams]

CUDA. That's all you need to know about nVidia's "willingness" to cooperate with others.

It cracked me when in the GTC Jen said "what makes ARM great is its openness". Like... Holy cow, how can you say that with a straight face, dude... No shame at all.

Cheers!

That situation could change due to the recent Supreme Court decision that disallowed copyrights of APIs. That opens the door to AMD and Intel supporting CUDA on their GPUs and, as a CUDA user myself, I really hope they do.

 

[-Fran-]

That situation could change due to the recent Supreme Court decision that disallowed copyrights of APIs. That opens the door to AMD and Intel supporting CUDA on their GPUs and, as a CUDA user myself, I really hope they do.

Oh... That is definitely news to me.

Do you have a link to share? That is absolutely interesting.

Cheers!

 

[InvalidError]

Oh... That is definitely news to me.

Do you have a link to share? That is absolutely interesting.

Oracle vs Google about the Java SE API:
https://www.supremecourt.gov/opinions/20pdf/18-956_d18f.pdf

Nvidia could still make trademark and possibly patent claims on CUDA though, so a 3rd-party re-implementation would at least need to be called something else due to not being authorized to use the name.

 

[Soaptrail]

That situation could change due to the recent Supreme Court decision that disallowed copyrights of APIs. That opens the door to AMD and Intel supporting CUDA on their GPUs and, as a CUDA user myself, I really hope they do.

Do Intel and AMD want to use CUDA? AMD has success with their own versions of Nvidia tech, e.g. Freesync.

 

[hotaru.hino]

Do Intel and AMD want to use CUDA? AMD has success with their own versions of Nvidia tech, e.g. Freesync.

No, because OpenCL is a thing. The problem though is NVIDIA had such a long headstart with the GPGPU idea that CUDA is entrenched in a lot of HPC software. So OpenCL hasn't exactly dominated over CUDA.

That situation could change due to the recent Supreme Court decision that disallowed copyrights of APIs. That opens the door to AMD and Intel supporting CUDA on their GPUs and, as a CUDA user myself, I really hope they do.

The ruling they did means that the interface itself isn't subject to copyright laws (or something). However, the implementation of said API is. So the ruling doesn't force NVIDIA to open up how they did CUDA. It just makes it so NVIDIA can't sue Intel or AMD for making their own implementation.

Which, Intel and AMD can make their own implementation anyway. It's what Google did with the Java Virtual Machine and why Oracle is mad.

 

[Gomez Addams]

Do Intel and AMD want to use CUDA? AMD has success with their own versions of Nvidia tech, e.g. Freesync.

That remains to be seen. Given the enormous amount of software written using CUDA I would think they do. As a user of CUDA I hope they do so I have more options available, hopefully at a bit lower cost. Currently Nvidia has no competitors in the CUDA-driven HPC market so they charge what ever they can get away with. I would like to see competition and lower prices.

Actually, my impression is they DO want to support CUDA. There was an unsupported version of CUDA that worked on AMD cards a while ago.

 

[Gomez Addams]

No, because OpenCL is a thing. The problem though is NVIDIA had such a long headstart with the GPGPU idea that CUDA is entrenched in a lot of HPC software. So OpenCL hasn't exactly dominated over CUDA.


The ruling they did means that the interface itself isn't subject to copyright laws (or something). However, the implementation of said API is. So the ruling doesn't force NVIDIA to open up how they did CUDA. It just makes it so NVIDIA can't sue Intel or AMD for making their own implementation.

Which, Intel and AMD can make their own implementation anyway. It's what Google did with the Java Virtual Machine and why Oracle is mad.

Duh. They always COULD make their own implementations. They essentially have to for their video cards to work with OS drivers. That is REQUIRED. The issue is making it source-code compatible with the CUDA API and now they can because of that ruling. That is what eliminating copyrights of APIs has enabled.

It isn't necessary for Nvidia to open up anything to see how CUDA works. CUDA code compiles to an intermediate, assembly-style format and anyone can look at that. The really tricky parts are hidden deeper and everyone who makes GPUs has their own proprietary mechanisms for those. They involve thread scheduling in computation units (Nvidia's SM) and higher level constructs like thread and block synchronization.

 

[hotaru.hino]

It isn't necessary for Nvidia to open up anything to see how CUDA works. CUDA code compiles to an intermediate, assembly-style format and anyone can look at that. The really tricky parts are hidden deeper and everyone who makes GPUs has their own proprietary mechanisms for those. They involve thread scheduling in computation units (Nvidia's SM) and higher level constructs like thread and block synchronization.

I'm pretty sure Dalvik wasn't some black-box reverse engineered copy of the JVM, but it's own thing that was compatible with Java. Especially when you consider JVM and Dalvik used different virtual machine designs. Which again, Intel and AMD could do on their own with CUDA and not care what NVIDIA does because at the end of the day, the API is just a black box as far as the application is concerned and as long as you get the expected output for a given input, it doesn't matter what goes between those steps.

But OpenCL exists, so why bother trying to make a re-implementation of CUDA?

 

[InvalidError]

But OpenCL exists, so why bother trying to make a re-implementation of CUDA?

The same reason any software or hardware vendor would adopt someone else's API: accessing their competitor's otherwise exclusive market.

 

[Gomez Addams]

....

But OpenCL exists, so why bother trying to make a re-implementation of CUDA?

Because performance matters.