Nvidia has approached Softbank and is considering buying ARM Holdings

Page 6 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

uzzi38

Platinum Member
Oct 16, 2019
2,565
5,574
146
Will update with articles as they come out (they haven't yet but news is public). For now just a couple of Tweets:



This is a real significant move for Nvidia, having a huge role in ARM core designs could have major implementations down the line. Most importantly, it secures them a platform and ecosystem down the line.

Oh and I guess Nvidia becomes the defacto standard for GPU IP for mobile instead of Mali. That too.

EDIT: Bloomberg article here: https://www.bloomberg.com/news/arti...ompany-arm-is-said-to-attract-nvidia-interest
 
Last edited:

Thala

Golden Member
Nov 12, 2014
1,355
653
136
Apple already has full control over the design of their CPUs. The only control ARM has is adding new instructions, and I'm sure Apple carries some influence for that.

Indeed ARMv8.4-8.6A was heavily driven by Apple. It is therefore no surprise, that ARM themselfs do not offer ARMv8.6 compliant architectures - however Apple A14 is an ARMv8.6 design.
 

soresu

Platinum Member
Dec 19, 2014
2,617
1,812
136
Indeed ARMv8.4-8.6A was heavily driven by Apple. It is therefore no surprise, that ARM themselfs do not offer ARMv8.6 compliant architectures - however Apple A14 is an ARMv8.6 design.
Not only Apple - I'm pretty sure that one of them has a significant mitigation measure against the recent spectre/meltdown issues.
 

Roland00Address

Platinum Member
Dec 17, 2008
2,196
260
126
A lot of patents. Full control of the design of their cpu's which are based on ARM and they are planning to use for everything and are key to the success of this 2 trillion dollar company. The cost is also insignificant to a company the size of Apple.

1) So Apple has access to all the patents already by buying that forever license they got from an “architectural license”, and architecture licenses actually cost more than the generic arm cores like A76.
2)The architectural license allows Apple to modify their designs and not fully comply in a positive or negative with the instruction set, aka they can add or remove features, and since they also fully control the software stack with the OS, Apps, and compiler layers adding or removing features is easy for them. Yes it may stink for the developer for they will have to recompile apps but literally apple is the best of the business with this,
3) Buying arm is cheap on a physical money front or offering Apple shares. It may not be cheap in other domains such as a waste of technical talent and attention, anti-trust, etc. The people who are saying it is not worth Apples time are saying ARM is a distraction, it does not add security in a form of a moat, it does not add value to Apple, and it wastes time and energy where attentive talent need to be focused on two dozen other things.
 

Doug S

Platinum Member
Feb 8, 2020
2,201
3,405
136
1) So Apple has access to all the patents already by buying that forever license they got from an “architectural license”, and architecture licenses actually cost more than the generic arm cores like A76.
2)The architectural license allows Apple to modify their designs and not fully comply in a positive or negative with the instruction set, aka they can add or remove features, and since they also fully control the software stack with the OS, Apps, and compiler layers adding or removing features is easy for them. Yes it may stink for the developer for they will have to recompile apps but literally apple is the best of the business with this,
3) Buying arm is cheap on a physical money front or offering Apple shares. It may not be cheap in other domains such as a waste of technical talent and attention, anti-trust, etc. The people who are saying it is not worth Apples time are saying ARM is a distraction, it does not add security in a form of a moat, it does not add value to Apple, and it wastes time and energy where attentive talent need to be focused on two dozen other things.

I don't believe #2 is correct. As I understand it, architectural licensees must implement the full ARMv8 ISA. They have a choice whether to implement the optional portions of ARMv8. If they want to implement a point extension like ARMv8.6 it is all or nothing as far as the required portion of that point release, and all previous point releases. I doubt they can add any new instructions either - if they added a new instruction using an encoding that ARM used in a future addition to the ISA it would create big problems.

This isn't really a hassle, they have to implement ISA instructions/features that are mandatory but there is nothing that says they have to use them. Not being able to add instructions wouldn't be something they are likely to care about either, since if there is an instruction useful to them it is probably useful to others and Apple would be able to convince ARM to include it in a future point release of the ISA. I saw a claim elsewhere that the A14 implements ARMv8.6, whereas the A12 and A13 implement ARM8.4. If so, presumably there is something added in .6 that they consider important to them but they didn't care about what .5 brought to the table (at least not enough to include it in the A13)

For example, if Apple decided to implement SVE2 on the A14 and thus in all CPUs used in the Mac, their compiler would generate instructions to always use SVE2 and never generate NEON instructions. They would still have to implement NEON on Mac CPUs, but nothing says it would have to be a good implementation - they could do a space efficient (which implies not performance efficient) implementation if they so chose (but probably wouldn't, at least not anytime soon, given Apple's stated support for Macs to run iOS applications, some of which will make use of NEON)

The reason they were able to drop ARMv7 support from their CPUs entirely is because when ARM defined ARMv8 they said it could come in three forms. 1) full implementation of ARMv8 and ARMv7, 2) full implementation of ARMv8 and user mode only implementation of ARM7, and 3) full implementation of ARMv8 only.

Allowing licensees to pick and choose what they implement, or add their own stuff willy nilly is not in the long term interest of the ARM ISA. It would not only be bad for ARM, it would be bad for licensees as they'd see a fracturing of tools like compilers that would need to have switches to generate "Apple code" and "Nuvia code" and so forth instead of a switch for ARMv8 and maybe --feature-sve2 or whatever.
 

NTMBK

Lifer
Nov 14, 2011
10,208
4,940
136
NVIDIA denies rumors of acquiring ARM


I just read the transcript, and no, JHH didn't deny anything. He just says "we've worked closely with ARM". That article is reading way too much into a non-answer.
 

mxnerd

Diamond Member
Jul 6, 2007
6,799
1,100
126
I just read the transcript, and no, JHH didn't deny anything. He just says "we've worked closely with ARM". That article is reading way too much into a non-answer.
Jensen Huang said that so far, NVIDIA has had a lot of cooperation with Arm, but has not yet proposed an acquisition.
 

NostaSeronx

Diamond Member
Sep 18, 2011
3,683
1,218
136
"We've been looking to be bought by ARM for a long time, and we use ARM in a whole bunch of applications. And whether it's autonomous driving or a robotics application, the Nintendo Switch, console business that we're in. And then, recently, we brought CUDA to ARM and to bring accelerated computing to ARM. And so, we hope to assimilate with the ARM team very soon.

They're really great guys. And one of the specials about the ARM architecture that you know very well is that it's incredibly energy-efficient. And because it's energy-efficient, it has the headroom to scale into very high-performance levels over time. And so, anyways, we love to join into one company with the ARM guys. Of course, with me at command!" - the Nosta cut

:laughing:
 

plopke

Senior member
Jan 26, 2010
238
74
101
Do you all believe from a market regulation this should even be allowed? Not that I have faith in any market regulation organization any more.

ARM is involved in so many industries and with direct NVIDIA competitors , so many things could just go horribly wrong.

Best case scenario they would get some knowledge of their competitors , how do the ARM engineers even talk to their customers what they would like to see changed/fixed with Jensen looking over your shoulder. He is saying it himself how much integration they have with ARM, so this can also be applied to their competitors?

Worst case NVIDIA could do like a entire Oracle
-Start manipulating service costs to shut down start ups , because lets be honest NVIDIA does never sell stuff for "cheap" where ARM has created services to make it easier to start producing ARM IP based products.
-Stop developing/improving parts that they not interested in anymore and do the bare minimum to keep contracts legal
-Forcing Nvidia IP on companies or adjusting stuff so it optimal to be only used with their software stack.
-Simply just ending contracts of competitors
-.....

And if they would do the worst case scenario and get sued for it , it would involve long trials in a market where before anything gets resolved it would be too late.
 
  • Like
Reactions: KompuKare

moinmoin

Diamond Member
Jun 1, 2017
4,933
7,619
136
@kokhua shared following insightful article on Twitter about Nvidia's interest in Arm. It does a good job describing the status quo (CUDA dominance) and what Nvidia would want to do to preserve it in face of Intel's and AMD's efforts, and how Arm comes into play.
 

Doug S

Platinum Member
Feb 8, 2020
2,201
3,405
136
@kokhua shared following insightful article on Twitter about Nvidia's interest in Arm. It does a good job describing the status quo (CUDA dominance) and what Nvidia would want to do to preserve it in face of Intel's and AMD's efforts, and how Arm comes into play.

Not sure I agree with analysis since it doesn't explain why Nvidia owning ARM is worth spending tens of billions, versus exercising an ARM architectural license to design their own custom ARM cores.

He talks about Nvidia needing to be completely vertically integrated. Well you know who else loves being completely vertically integrated? Apple. If Apple is satisfied designing their own custom cores gives them what they need to be fully vertically integrated without the disadvantages (spending money to buy a business they don't want to have to operate, regulatory issues, etc.) I fail to see how Nvidia's situation in the datacenter is sufficiently different that they should feel that's not good enough.
 

moinmoin

Diamond Member
Jun 1, 2017
4,933
7,619
136
Not sure I agree with analysis since it doesn't explain why Nvidia owning ARM is worth spending tens of billions, versus exercising an ARM architectural license to design their own custom ARM cores.

He talks about Nvidia needing to be completely vertically integrated. Well you know who else loves being completely vertically integrated? Apple. If Apple is satisfied designing their own custom cores gives them what they need to be fully vertically integrated without the disadvantages (spending money to buy a business they don't want to have to operate, regulatory issues, etc.) I fail to see how Nvidia's situation in the datacenter is sufficiently different that they should feel that's not good enough.
Apple already had and still has completely vertically integrated thanks to its software (though one could argue moving away from Intel allows them to put an end to unofficial Mac clones, extending it some more), it doesn't need exclusive access to Arm for that. That's exactly where Nvidia wants to get. Currently Nvidia's dominance in the datacenter AI/ML is thanks to CUDA so based on software alone, that won't last. Nvidia still relies on 3rd party CPU manufactures as the last leg of its three-legged chair which makes the software more interchangeable, and both Intel and AMD are pushing into that market with more open standards. Nvidia's market valuation is big enough that buying Arm through a partly stock deal is feasible. And hijacking and slowly subverting the Arm ecosystem would allow Nvidia both better vertical integration and a boarder market being locked into it.

It's a very risky devil's advocate approach, imo not unthinkable of Huang after his successful buyout of Mellanox (a multi-billion deal itself) that turned out to pay off much faster than expected.
 

DrMrLordX

Lifer
Apr 27, 2000
21,582
10,785
136
Not sure I agree with analysis since it doesn't explain why Nvidia owning ARM is worth spending tens of billions, versus exercising an ARM architectural license to design their own custom ARM cores.

He's using a lot of words to basically say: NV needs control of the entire datacentre platform, and ARM gives them the CPU + motherboard/chipset control that NV doesn't have right now. It's basically a bet that ARM can upend x86 and bypass POWER in the server and HPC space. Which is where NV is already selling a lot of their high-end accelerators already. NV gets to effectively replace Marvell, Amazon, Fujitsu, Huawei, and anyone else trying to produce ARM-based server hardware. As an added bonus you get their CUDA-based accelerators and their Mellanox-based interconnects thrown in as a package deal.

Plus they can kill Mali and put their own iGPUs in phones/tablets/etc.
 

Doug S

Platinum Member
Feb 8, 2020
2,201
3,405
136
He's using a lot of words to basically say: NV needs control of the entire datacentre platform, and ARM gives them the CPU + motherboard/chipset control that NV doesn't have right now. It's basically a bet that ARM can upend x86 and bypass POWER in the server and HPC space. Which is where NV is already selling a lot of their high-end accelerators already. NV gets to effectively replace Marvell, Amazon, Fujitsu, Huawei, and anyone else trying to produce ARM-based server hardware. As an added bonus you get their CUDA-based accelerators and their Mellanox-based interconnects thrown in as a package deal.

Plus they can kill Mali and put their own iGPUs in phones/tablets/etc.

I don't see how owning ARM helps Nvidia upend x86 and bypass POWER any better than being an architectural licensee does. The competitors in the ARM server space they need to worry about are also architectural licensees, so they won't be able to kill them off.

Still not buying this argument until I see someone come up with a reasonable explanation for the advantage ownership of ARM confers to Nvidia beyond what they get as an architectural licensee.
 
  • Like
Reactions: Tlh97

ThatBuzzkiller

Golden Member
Nov 14, 2014
1,120
260
136
I don't see how owning ARM helps Nvidia upend x86 and bypass POWER any better than being an architectural licensee does. The competitors in the ARM server space they need to worry about are also architectural licensees, so they won't be able to kill them off.

Still not buying this argument until I see someone come up with a reasonable explanation for the advantage ownership of ARM confers to Nvidia beyond what they get as an architectural licensee.

Isn't that obvious ? If they buy ARM Ltd they get to make more rules for their own benefit so that other players are forced to take them ...

By using ARM as a leverage point against other vendors they can also attempt to promote further lock-in towards their platform by deprecating Mali GPUs and forcing other vendors who don't have in-house graphics hardware design teams to be reliant Nvidia graphics IPs ...

Little do Nvidia know that it's going to be met with limited success. I guess buying out Qualcomm was too pricey for them ...
 

Thala

Golden Member
Nov 12, 2014
1,355
653
136
Isn't that obvious ? If they buy ARM Ltd they get to make more rules for their own benefit so that other players are forced to take them ...

I dont see how this would help them.

By using ARM as a leverage point against other vendors they can also attempt to promote further lock-in towards their platform by deprecating Mali GPUs and forcing other vendors who don't have in-house graphics hardware design teams to be reliant Nvidia graphics IPs ...

They just need a better GPU than MALI for the mobile space - but for this they do not need ARM either.
 
  • Like
Reactions: Tlh97

Doug S

Platinum Member
Feb 8, 2020
2,201
3,405
136
Isn't that obvious ? If they buy ARM Ltd they get to make more rules for their own benefit so that other players are forced to take them ...

By using ARM as a leverage point against other vendors they can also attempt to promote further lock-in towards their platform by deprecating Mali GPUs and forcing other vendors who don't have in-house graphics hardware design teams to be reliant Nvidia graphics IPs ...

Little do Nvidia know that it's going to be met with limited success. I guess buying out Qualcomm was too pricey for them ...

What rules can they make that would hurt architectural licensees? They can't rewrite existing contracts. Sure, they can hurt Qualcomm, Samsung, and Huawei who currently use licensed cores by withholding future cores or raising the price excessively, but they all have architectural licenses so they can either resume designing their own cores and ignore the new owner of ARM like Apple would be, or being making plans to abandon ARM long term for RISC-V or whatever.

If Nvidia is doing this to help their position in the datacenter world, hurting Qualcomm and Samsung does nothing to accomplish that goal. The best they could hope for without killing the goose would be offering prices similar to the current ones for ARM cores (and a long term license to alleviate FUD and keep them from restarting their own architecture teams) if they agree to use Nvidia GPUs.

I think the "trying to force the Android world to switch to Nvidia GPUs by leveraging ARM" might be the only thing that makes sense, as it would presumably help them make inroads into the PC marketplace (especially if Windows/ARM ever gains any traction) I think it is a huge stretch to say that's worth $30 billion or whatever, but who knows.

That idea at least has possibilities. The idea they are doing it for datacenter has none, and makes absolutely no sense when viewed from any angle.
 
  • Like
Reactions: Tlh97

Vattila

Senior member
Oct 22, 2004
799
1,351
136
Still not buying this argument until I see someone come up with a reasonable explanation for the advantage ownership of ARM confers to Nvidia beyond what they get as an architectural licensee.

Agree. I find it hard to see why Nvidia should want to become a licensing house for ARM IP. That has too many complicating factors. I do however see the attraction in acquiring the operational side of ARM — the actual world class CPU design team and access to the IP they have built up. They have an impressive track record and a great roadmap for entering the server space. Jensen would love to have that, so that he could hit the ground running in his ambitions to become a competitor in the server CPU market, and hence play a fuller role in the HPC space. In time, that CPU capability may trickle down to the high margin workstation and HEDT segments as well.

If an acquisition goes through, I guess the IP licensing business, and associated patent portfolio will likely end up controlled by a consortium, or retained by SoftBank.
 

Doug S

Platinum Member
Feb 8, 2020
2,201
3,405
136
Agree. I find it hard to see why Nvidia should want to become a licensing house for ARM IP. That has too many complicating factors. I do however see the attraction in acquiring the operational side of ARM — the actual world class CPU design team and access to the IP they have built up. They have an impressive track record and a great roadmap for entering the server space. Jensen would love to have that, so that he could hit the ground running in his ambitions to become a competitor in the server CPU market, and hence play a fuller role in the HPC space. In time, that CPU capability may trickle down to the high margin workstation and HEDT segments as well.

If an acquisition goes through, I guess the IP licensing business, and associated patent portfolio will likely end up controlled by a consortium, or retained by SoftBank.

I would assume Nvidia has a pretty good design team, given their success in the GPU and HPC markets.

It is a lot cheaper to hire away ARM and other company's top designers with huge salaries than to buy a company for $30 billion to get them (and hope they are still there once you take over)
 

Vattila

Senior member
Oct 22, 2004
799
1,351
136
I would assume Nvidia has a pretty good design team, given their success in the GPU and HPC markets.

It is a lot cheaper to hire away ARM and other company's top designers with huge salaries than to buy a company for $30 billion to get them (and hope they are still there once you take over)

True. But acquiring a well-functioning and highly performing team intact has its attractions. The existing GPU designers at Nvidia may have their hands full already, even if they should prove to be equally capable of CPU design, which by the way is far from given.

On the shape of the acquisition, I doubt Nvidia will pay $30B. That valuation has been put on ARM as a whole. The CPU design team and associated IP may be a fraction of that, assuming the licensing business is spun off into a consortium or retained by SoftBank. Likewise, I guess the Mali GPU team will be retained, spun off or dismantled, as presumably it is of little interest to Nvidia (if they are interested in some of their GPU designers, they can attempt to lure them away, as you suggest), although some of the Mali GPU IP may be attractive, for which they may seek a perpetual license.
 

DrMrLordX

Lifer
Apr 27, 2000
21,582
10,785
136
I don't see how owning ARM helps Nvidia upend x86 and bypass POWER any better than being an architectural licensee does. The competitors in the ARM server space they need to worry about are also architectural licensees, so they won't be able to kill them off.

NV can probably soft-terminate architectural licenses by modifying the ISA in such a way that it is no longer covered in the license. Also you will note that almost all the ARM hardware vendors are pushing cores that are based on ARM reference designs (exceptions being Fujitsu and Huawei). Even if the architectural licenses do allow Marvell/Fujitsu/Amazon/Huawei/etc. to continue producing CPUs in compliance with some version of the ARM ISA, they'll be stuck producing their own designs from the ground up instead of using something like Matterhorn, or whatever else. And again, they may be trapped supporting an "old" version of the ISA.

This assumes that ARM can supplant x86 and put the final nails in POWER's coffin, which is . . . not certain at this point.

But we're getting away from the central point.

nVidia needs platform control. Period. If they continue on as they are, Intel and AMD will try to deplatform them. There may be no future for nVidia on competing non-x86 server platforms either - in fact, that's a big question mark. Fujitsu seems to be betting that compute accelerator cards are not the way to go (A64FX is moving in that direction), and if/when SVE2 is embraced broadly by companies like Marvell or Huawei, they may move in that direction as well.
 
Last edited:
  • Like
Reactions: Tlh97 and moinmoin

NTMBK

Lifer
Nov 14, 2011
10,208
4,940
136
True. But acquiring a well-functioning and highly performing team intact has its attractions. The existing GPU designers at Nvidia may have their hands full already, even if they should prove to be equally capable of CPU design, which by the way is far from given.

NVidia already has a CPU design team- they created the Denver, Denver2 and Carmel CPU cores.
 
  • Like
Reactions: Tlh97 and Vattila

Vattila

Senior member
Oct 22, 2004
799
1,351
136
NVidia already has a CPU design team- they created the Denver, Denver2 and Carmel CPU cores.

True. I guess the idea is to strengthen that team. Otherwise, if Jensen feels he already has the team and IP needed to win in the server CPU space, I don't see why he is wasting time and money on these alleged acquisition talks with SoftBank. The idea that he aims to control ARM to sabotage his competitors is for the birds.