Nvidia has approached Softbank and is considering buying ARM Holdings

Page 2 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

VirtualLarry

Lifer
Aug 25, 2001
48,384
5,094
126
Honestly, I see an NVidia-ARM merger/purchase as a pretty incredible tech and business synergy, to use "corporate-speak".

It would finally give NVidia a decent CPU architecture base (although, don't they already have an ARM license?), and by bundling NVidia GPU tech, they could make it the de-facto standard, at least in mobile, in the next few years, and they could expand to desktop PCs, if Apple is doing it with ARM-based CPUs, then certainly, NVidia can make a go of it with OEMs, if they're willing. (Hint, Nvidia might give them a sort of quid-pro-quo "push" to build NVidia-ARM-based laptops, if those laptop OEMs want access to NVidia's dGPU product lineup for their Intel CPU-based laptops. "Informally", of course.)
 
  • Like
Reactions: Tlh97 and CHADBOGA

soresu

Golden Member
Dec 19, 2014
1,324
520
136
From NVs point of view this makes sense as AMD and/or intel could push them out of the market over time by making their card incompatible with their CPUs so their need their own.
nVidia getting squeezed out isn't going to happen.

I think that there were some issues back in the 00's on some AMD chipsets, but that was a long time ago now, it's not in anybodies interests to do that.

The moment Intel or AMD try to lockdown their platform to only their own products, their competitors will takes advantage by proclaiming themselves the open choice allowing for free market competition.

nVidia would be as likely to make a high end ARM system of their own and make it so that only their discrete GPU products work on it - it just ends up restricting your potential customer base.
Amazon buying actually makes a ton of sense, and they could get it for free by issuing stock.
Definitely, they started with SoC's for NAS boxes and now AWS so it's certainly in their wheelhouse.

Their N1 based investments show that they see the merits of ARM's silicon so they would be unlikely to stick their oar in and try to change direction.
 

Roland00Address

Golden Member
Dec 17, 2008
1,929
64
91
It's not old school yet, not even remotely.

It's still state of the art until there is at least 1, or even 2 generations of v9-A cores both big and little on the market.

Even then, I believe that v8-A code, at least v8-A A64 will remain compatible in the v9-A ISA based cores, much as v7-A is still supported by v8-A now

But I believe that v7-A (A32) support will finally be deprecated in hardware at least for v9-A, probably relegated to software emulation for the few apps that have lazy developers who haven't already made the jump to v8-A A64 backends.

Both Apple first and then Google later were keen to enforce new apps and app updates be A64 only within the last few years, and I believe that this was at least part of the reason beyond platform tech support headaches.
We are getting in the weeds now since you took offense. I am not going to get rough and tussle about this.

But if you are making a new soc today, something that takes 2 to 4 years to get from design to final product you will try not to use v8 as the base and instead use v9.

Furthermore while we can not prove it for the sake of the thread it is likely ARM partners have already seen some of V9 right now but we can not prove it.

Sure some ARM v8 stuff that was designed in the last 4 years (planning stage here is what I am referring to, not the last 4 years of shipping stuff) will be competing with ARM v9. Regardless the instruction set is old school and was announced in 2011* and apple shipping their silicon in 2013 with the A7.

*Yes that was the Arm v8-A there has been 6 revisions / additions since then with the current best hotness instruction wise is the 2019 Arm v8.6-A.
 

soresu

Golden Member
Dec 19, 2014
1,324
520
136
We are getting in the weeds now since you took offense. I am not going to get rough and tussle about this.

But if you are making a new soc today, something that takes 2 to 4 years to get from design to final product you will try not to use v8 as the base and instead use v9.

Furthermore while we can not prove it for the sake of the thread it is likely ARM partners have already seen some of V9 right now but we can not prove it.

Sure some ARM v8 stuff that was designed in the last 4 years (planning stage here is what I am referring to, not the last 4 years of shipping stuff) will be competing with ARM v9. Regardless the instruction set is old school and was announced in 2011* and apple shipping their silicon in 2013 with the A7.

*Yes that was the Arm v8-A there has been 6 revisions / additions since then with the current best hotness instruction wise is the 2019 Arm v8.6-A.
It was not a matter of taking offense, though clearly you took my reply as such for some strange reason of your own.

None of your reply has a shred of meaning as it is completely speculative given the reality of the now vs your as yet unrealised future.

ARM has not even officially announced v9-A yet, and you are getting stroppy about someone telling you that the actual v8-A products on the market represent the state of the art?

When I say state of the art, I mean things that are actually verified information like Matterhorn's 10x ML performance from MatMul - and not speculations of the forum which are all that exist about v9-A at the moment besides job postings.

Also to be clear, by v8-A I meant the entire gamut of v8.x-A based products.

It's a pain to go into exact detail on exact v8.x-A spec support of any given product because exact ISA spec support is not at all consistent across the market currently.

ARM have been on a sort of v8.2-A+ spec since A75 and A55 as they also support some features of post v8.2-A specs, but not enough to call it a full step up from the 2017 iSA release.

Even Apple only supports v8.4-A at this point.
 

ThatBuzzkiller

Senior member
Nov 14, 2014
956
125
106
Honestly, I see an NVidia-ARM merger/purchase as a pretty incredible tech and business synergy, to use "corporate-speak".

It would finally give NVidia a decent CPU architecture base (although, don't they already have an ARM license?), and by bundling NVidia GPU tech, they could make it the de-facto standard, at least in mobile, in the next few years, and they could expand to desktop PCs, if Apple is doing it with ARM-based CPUs, then certainly, NVidia can make a go of it with OEMs, if they're willing. (Hint, Nvidia might give them a sort of quid-pro-quo "push" to build NVidia-ARM-based laptops, if those laptop OEMs want access to NVidia's dGPU product lineup for their Intel CPU-based laptops. "Informally", of course.)
I highly doubt Nvidia SoCs would become the defacto standard for mobile devices. They are missing a key technology like wireless connectivity. If ARM does go on to design exclusively for Nvidia after the acquisition then it's more likely that either Huawei, Qualcomm, Samsung or even Apple will be more likely to come out on top. Good graphics technology isn't worth giving up support for modern wireless network technology like 4G LTE or 5G NR ...

Tegra sank because Nvidia couldn't even figure out LTE technology. To this day, getting a Tegra is like asking if you want to be blacklisted from every 5G mobile carrier network ...
 
  • Like
Reactions: Tlh97

soresu

Golden Member
Dec 19, 2014
1,324
520
136
Tegra sank because Nvidia couldn't even figure out LTE technology. To this day, getting a Tegra is like asking if you want to be blacklisted from every 5G mobile carrier network ...
*Tegra pivoted, it didn't sink - yet.

Even though TX1 is relatively old in market terms there are still millions of Switch consoles getting sold.

The real question is whether there will be an entirely new Tegra chip for Switch 2 or if they will change vendors.

For certain if they made a new SHIELD TV with even an A76 CPU and just 3-4x the Tegra X1 GPU grunt I would definitely buy it, the 'competitors' in the Android TV market need a kick in the posterior to get them really competitive for once.
 

ThatBuzzkiller

Senior member
Nov 14, 2014
956
125
106
*Tegra pivoted, it didn't sink - yet.

Even though TX1 is relatively old in market terms there are still millions of Switch consoles getting sold.

The real question is whether there will be an entirely new Tegra chip for Switch 2 or if they will change vendors.

For certain if they made a new SHIELD TV with even an A76 CPU and just 3-4x the Tegra X1 GPU grunt I would definitely buy it, the 'competitors' in the Android TV market need a kick in the posterior to get them really competitive for once.
Tegra is pretty much a dead end. The Switch being mildly successful just happens to be a coincidence and it doesn't make a meaningful impact in Nvidia's pockets ...

I bet Nvidia wish it could capitalize a large portion of the mobile devices out there like Qualcomm does instead of the Switch. Portable gaming systems are a drop in the bucket compared to handsets ...

SHIELD TV hardware was pretty much useless without good software unlike the Switch so nothing will change even if Android TV device manufacturers cared because they aren't going to be making high-end games. Good hardware alone isn't going to sell itself no matter how much you force yourself to like this model ...
 
  • Like
Reactions: Tlh97 and Summerfun

soresu

Golden Member
Dec 19, 2014
1,324
520
136
Tegra is pretty much a dead end. The Switch being mildly successful just happens to be a coincidence and it doesn't make a meaningful impact in Nvidia's pockets ...

I bet Nvidia wish it could capitalize a large portion of the mobile devices out there like Qualcomm does instead of the Switch. Portable gaming systems are a drop in the bucket compared to handsets ...

SHIELD TV hardware was pretty much useless without good software unlike the Switch so nothing will change even if Android TV device manufacturers cared because they aren't going to be making high-end games. Good hardware alone isn't going to sell itself no matter how much you force yourself to like this model ...
Yah no.

First off Switch is not "mildly successful", it is very successful by any business terminology at over 55 million units shipped since just 2017.

It's even more successful still as their R&D investment in developing the console can't have been that much, even compared to the post Gamecube weaksauce hardware revamps.

Second I wasn't implying SHIELD TV are really any good for ordinary Android games - but for emulators and being a proper decent streamer it still competes even with its ancient CPU, because much as I hate to admit it (as I really do hate nVidia) they have supported the STV software uncommonly well, certainly far better than most smartphone manufacturers do their products.

I don't know what their deal with Nintendo is, but all things considered it's a great one for nVidia given how little effort they have to put in to get a return - it's not like TX1 was even a Volta/Turing size chip to begin with, so even shrinking it to 16nm can't have been such a chore.

Whoever got that deal sealed with Nintedo got a promotion and then some mark my words.
 

ThatBuzzkiller

Senior member
Nov 14, 2014
956
125
106
Yah no.

First off Switch is not "mildly successful", it is very successful by any business terminology at over 55 million units shipped since just 2017.
Whatever floats your boat and Nvidia only gets to see a tiny fraction of that success on their side compared to Qualcomm raking in billions since they supply a good portion of the modems and antenna modules for the high-end smartphone market.

Nvidia only wishes that their hardware was actually valuable enough in that segment to achieve a fraction of what Qualcomm and the other were able to ...

Second I wasn't implying SHIELD TV are really any good for ordinary Android games - but for emulators and being a proper decent streamer it still competes even with its ancient CPU, because much as I hate to admit it (as I really do hate nVidia) they have supported the STV software uncommonly well, certainly far better than most smartphone manufacturers do their products.
Drivers and OS updates means little. Switch is successful because of it's games, not because Nvidia supplied the hardware ...

I don't know what their deal with Nintendo is, but all things considered it's a great one for nVidia given how little effort they have to put in to get a return - it's not like TX1 was even a Volta/Turing size chip to begin with, so even shrinking it to 16nm can't have been such a chore.

Whoever got that deal sealed with Nintedo got a promotion and then some mark my words.
Let's get back to the topic. The only hardware that matters in a mobile devices is the baseband processor and surprisingly enough the memory modules since Samsung has a near stranglehold domination with that specific component ...
 

soresu

Golden Member
Dec 19, 2014
1,324
520
136
Drivers and OS updates means little. Switch is successful because of it's games, not because Nvidia supplied the hardware ...
I didn't say that nVidia was the reason for the Switch's success - you said it was "mildly successful", I refuted the "mildly" part, not the why or how of its success.

And as someone who has repeatedly screamed at vendor support pages while waiting for the next Android update from companies that make millions, I can say that updates are not trivial - especially for me waiting quite some time to finally get dark mode from the Android 10 update.

Long time support matters to a customer not willing to get a new phone every year - especially now that the benefits of doing so are generally unnoticeable for basic operations.

On that note it's actually getting rather comical that they keep making new phones bigger and bigger each generation - it's not trivial to find a decent new phone that isn't enormous and difficult to fit in the pockets of normal people clothes.
Let's get back to the topic. The only hardware that matters in a mobile devices is the baseband processor and surprisingly enough the memory modules since Samsung has a near stranglehold domination with that specific component ...
Stranglehold?

What do no one else make LPDDRx modules or something?

That sounds like a price fixing situation more than any "stranglehold domination".

It seems like regulators aren't even trying to prevent it anymore, it basically killed the BDR drive by price fixing the blue violet laser diodes they use, that's the main reason that the few computers that still come with optical drives come with DVDR instead of BDR drives.

I can only assume that Sony and MS entered into an unholy alliance with the providers for the sake of the consoles that use them too.

*on the subject of dark mode, I don't suppose this forum has one yet?

My eyesight is bad and focusing on light text on dark backgrounds just seems to cause less headaches, so any help appreciated.
 
Last edited:

DrMrLordX

Lifer
Apr 27, 2000
16,023
4,970
136
If you can't beat em buy em.
I don't think nVidia has ever really tried to "beat" ARM holdings. But they have tried to compete with ARM licensees and failed (Tegra).

Tegra is pretty much a dead end. The Switch being mildly successful just happens to be a coincidence and it doesn't make a meaningful impact in Nvidia's pockets ...
The Switch's success is predicated upon Tegra being a dead-end. Nintendo did not want a good piece of hardware. They wanted a CHEAP piece of hardware that was still functional enough for their in-house software. Tegra is cheap because of how few people will buy it. Nintendo STILL hasn't adopted TX2. nVidia could continue to feed Tegra to Nintendo in the form of a die-shrunk/updated TX2 for Switch 2 (or whatever).
 
  • Like
Reactions: Tlh97

Lodix

Senior member
Jun 24, 2016
299
85
101
*on the subject of dark mode, I don't suppose this forum has one yet?

My eyesight is bad and focusing on light text on dark backgrounds just seems to cause less headaches, so any help appreciated.
OT: I use the app Tapatalk to enter the forum. It has a dark mode that is easy on the eyes and good for battery life on my OLED display.
 
  • Like
Reactions: Tlh97 and soresu

beginner99

Diamond Member
Jun 2, 2009
4,539
885
126
The moment Intel or AMD try to lockdown their platform to only their own products, their competitors will takes advantage by proclaiming themselves the open choice allowing for free market competition.
Well with NV, anyone is going to laugh about the open notion. Just look at CUDA. But that wouldn't be the problem NV still makes their large share of sales on the gaming market, the x86 gaming market. See the problem?
 

DisEnchantment

Senior member
Mar 3, 2017
578
1,077
106
Well with NV, anyone is going to laugh about the open notion. Just look at CUDA. But that wouldn't be the problem NV still makes their large share of sales on the gaming market, the x86 gaming market. See the problem?
In my professional experience, all we have ever got with Tegra deliveries are stripped libraries/binaries and header files. Good luck troubleshooting stuffs.
Our other suppliers deliver us code so that we can debug and adapt when something is wrong in our platforms.
Can't get any farther from open than NV.
 

Asterox

Senior member
May 15, 2012
333
326
136
In my professional experience, all we have ever got with Tegra deliveries are stripped libraries/binaries and header files. Good luck troubleshooting stuffs.
Our other suppliers deliver us code so that we can debug and adapt when something is wrong in our platforms.
Can't get any farther from open than NV.
And we all now what Linus Torvalds said about Nvidia. :relieved:

As goes for Nvidia maybe buy ARM, no it will not hapened. Nvidia does not have that kind of big $(not even close), the money was alredy spent on other purchases.

 

Ajay

Diamond Member
Jan 8, 2001
7,064
2,364
136
In my professional experience, all we have ever got with Tegra deliveries are stripped libraries/binaries and header files. Good luck troubleshooting stuffs.
Our other suppliers deliver us code so that we can debug and adapt when something is wrong in our platforms.
Can't get any farther from open than NV.
The don't even provide your company with libraries compiled with symbols on?
 

DisEnchantment

Senior member
Mar 3, 2017
578
1,077
106
The don't even provide your company with libraries compiled with symbols on?
Well... I don't want to write all those things on here. :confused:

But on OpenEmbedded Cross build framework, strip is a step which can be added before packaging for delivery to the System Integrator. So you are out of luck if you have a crash dump. Forget about source code.
 

dullard

Elite Member
May 21, 2001
22,325
690
126
As goes for Nvidia maybe buy ARM, no it will not hapened. Nvidia does not have that kind of big $(not even close), the money was alredy spent on other purchases.
Last I saw NVidia has $16 billion in cash, which is roughly half of what is needed. Insiders have $10B of stock. With trillions of dollars floating around with all the stimulus packages, getting a loan for the rest will be a piece of cake. Heck the US Fed said it would buy unlimited corporate debt. Alternatively, with $250B market cap, just do a stock trade or even issue a bit more stock. NVidia can easily purchase ARM.
 
  • Like
Reactions: HurleyBird

dark zero

Platinum Member
Jun 2, 2015
2,539
98
91
This could be nVidia reacting to the Samsung RDNA news.

nVidia won't want AMD succeeding where they failed in mobile, even if it's only through licensing their GPU IP to others.
Wait, nVIDIA getting ARM would end Mali permanently.
And that would heavily hurt Mediatek and UNISOC. Unless they are forced to go to Power VR or AMD
 

soresu

Golden Member
Dec 19, 2014
1,324
520
136
Well with NV, anyone is going to laugh about the open notion. Just look at CUDA.
I wasn't talking about CUDA software but rather nVidia card support on AMD/Intel platforms (basically all x86 ATX/ITX) which is what I presumed you were talking about.

Restricting their platforms would not play well, and both AMD and Intel know this.

The only modern example even close is the restriction of UHD Bluray playback to Intel specific systems with SGX instructions and their own iGPU - which is itself more on Cyberlink and the Bluray Disc Association than Intel themselves.
 

Roland00Address

Golden Member
Dec 17, 2008
1,929
64
91
It was not a matter of taking offense, though clearly you took my reply as such for some strange reason of your own.

None of your reply has a shred of meaning as it is completely speculative given the reality of the now vs your as yet unrealised future.
Yep I read you correctly the first time, and I will not be participating further. You know things (except you don't) while simultaneously discounting anything another person said for as we both agree the future is speculative.

Bye, bye to this conversation topic while still participating with the rest of this thread..

Tegra sank because Nvidia couldn't even figure out LTE technology. To this day, getting a Tegra is like asking if you want to be blacklisted from every 5G mobile carrier network ...
Tegra sank due to LTE, but even if they got it figured out the Tegra chips were just "weird" chips.

Not having LTE would sank any SoC, but my point here is Tegra was a small chip, missing key features that everyone thought was important, yet still have a decent gpu. Tegra was "designed" not for actual markets but instead what Nvidia hoped the market would do. Nvidia hoped the market would do the things that were Nivida's strengths (the GPU), Nvidia did little to adapt themselves to the existing marketing or where people forsaw the market going in the near future... and Nvidia paid for the price for that is not what the various OEM customers wanted. (With the exception of Automotive.) Seriously the limited design wins Nvidia did get were due to the fact the Tegra 2 was the reference design for Android 3.0 / Honeycomb, the first tablet OS for Android.
 

Doug S

Senior member
Feb 8, 2020
250
301
96
Last I saw NVidia has $16 billion in cash, which is roughly half of what is needed. Insiders have $10B of stock. With trillions of dollars floating around with all the stimulus packages, getting a loan for the rest will be a piece of cake. Heck the US Fed said it would buy unlimited corporate debt. Alternatively, with $250B market cap, just do a stock trade or even issue a bit more stock. NVidia can easily purchase ARM.
You're assuming Softbank can get anything like the $32 billion they paid. That's a pretty big price for a company with REVENUE of only $1.6 billion, and that already dominates licensable CPU cores to such an extent that future growth can't equal past growth (and free alternatives like RISC-V seriously limit their ability to collect much money in future low cost markets like IoT)
 

ASK THE COMMUNITY