Nvidia in 'Advanced Talks' to Acquire Arm

[Gerdi]

Apple doesn't need to license anything else with their current SoCs as those are under an existing ARM ISA license that is in force in perpetuity.

Only if ARM develops an all-new ISA (that I expect would have to be fundamentally different from the existing ARM ISA) that Apple would be interested in using to develop their own SoCs on would they then need to secure a new license.

But I expect this would be unlikely considering the huge installed base of devices that use the existing ARM ISA. It would be kind of like when Intel tried to move everyone from x86 to Itanium and we know how well that worked.

Apple's existing ARM ISA license cannot be terminated.

I bet, Apple would like to switch to ARMv9 once the world switches to ARMv9 - which requires an ARMv9 license.
In addition, there is a very strong partnership between Apple and ARM. You can clearly see, that most features of ARMv8.4-ARMv8.6 are driven by Apples requirements - and ARM just put them into the architecture specification.

 

[deconstruct60]

All I want is for Nvidia GPUs to come back to the Mac. Hopefully this helps make that happen.

Pretty good chance this will deepen , not lessen , the 'dust up' between Apple and Nvidia. Nvidia GPUs would be as unlikely state on macOS as ARM's GPU.

Apple has a license for the parts of the ARM IP they use now. If Nvidia gets heavy handed in cranking up licensing costs and/or trying to act like they have deep leverage on Apple then , Nvidia will just dig a deeper hole.

Yet another substantive income stream means Nvidia "needs" the small fragment of dGPUs left of the macOS even less. That doesn't put Apple in a better leverage position either.

Apple is out to remove as many discrete GPUs as that can from most of the Mac line up. Metal is top priority and anyone who can't get on board with that is 'out'. ( as long as it is "CUDA first and Metal second' ... not going to get in. ) . Metal will just be even more deeply entrenched on macOS when Apple GPU is the largest volume GPU on Macs ( versus Intel iGPU now) .


Even if Nvidia runs ARM has a "hands off" , wholly own subsidiary where doesn't get looped into the Nvidia - Apple 'dust up'. That won't help resolve that situation either.

ARM's limitation of being IP licensing business doesn't particularly look promising for its long term prospect.

ARM still gets royalties per unit. Look term prospects are good. ( traction in the sever market is about to get substantial). Matching highly speculative ( edge of irrational exuberance ) valuation long term might be a problem. But it isn't like ARM shouldn't make money over the long term. Whether they can make giant buckets of profit long term is a different issue.

I think SoftBank should do an IPO for ARM. In this funny money environment that we are in, ARM will easily be valued at more than the $32B that they paid for it and they won't have to worry about any regulatory hassles.

The big problem with an IPO is far more so that about zero of that money would go to ARM for long term investment. Softbank will siphon off almost all of the cash they can legally walk away with. ARM completely stripped of any investment resources would not be good ( or a stable position. )

An IPO has a pretty good chance of leaving ARM close to 'broke' . That isn't what an IPO should do.

The sell here of ARM is because Softbank is eyeball deep in trouble. ARM is the one of the larger investments they made that might possibly get a return on if Softbank can squeeze it dry and move on.


ARM sold to someone that has deeper pockets to continue to do robust forward investment would be far more likely to lead to good outcomes later for ARM.

 

[370zulu]

Not excited about this news one bit.

shaka, when the walls fell.

 

[JosephAW]

Does this mean that an ARM hackintosh could happen?

 

[Analog Kid]

Engineers 😅🤷‍♂️

Whenever there was some new feature being worked on (e.g. some new video encoding tech or whatever), the company would be shopping around for licensees who might be interested. From what I heard, Apple was basically never interested - they just did their own thing. Their chips are technically “ARM cores“, but they are nothing like ARM’s own designs. Really, they’ve been “Apple cores” for a long time already.

I don’t know any of the specifics, because ARM is very tight on information security, and it was roughly 8 or so years ago anyway.

Which I think is why they keep hammering home "Apple Silicon" rather than Mac on ARM. I don't think Apple much cares what happens to ARM at this point, they've got their own destiny. If there was ever a risk that anyone could revoke Apple's right to keep making chips, there's no way they would have stepped into the breach.

 

[CWallace]

I bet, Apple would like to switch to ARMv9 once the world switches to ARMv9 - which requires an ARMv9 license. In addition, there is a very strong partnership between Apple and ARM. You can clearly see, that most features of ARMv8.4-ARMv8.6 are driven by Apples requirements - and ARM just put them into the architecture specification.

It is possible Apple has already secured an ARMv9 ISA license. Annandtech's article on the Apple Silicon announcement was commenting on whether the first generation of Apple Silicon SoCs would be using v8.2 and its NEON extensions or if it could be the first ARMv9 SoC.

 

[Analog Kid]

Once again, nothing fundamentally wrong with the x86 ISA. People are confusing Intel's execution problem with the x86 ISA.

If Intel wanted to optimize for power and die size, Intel could strip the 16- and 32-bit features from x86, the way Apple did with their implementation. But x86 is all about backwards compatibility, while Apple only cares about the last 5 years.

I don't think you can really say "there's nothing wrong with the architecture except for all the cruft". Backwards compability is the x86 ISA.

If the implication is that Intel chose not to sell into the highest growth market ever because they just don't like money, then the problems at Intel are worse that I thought and AMD missed a huge opportunity.



Intel didn't want a piece of that, or couldn't get a piece of that?

 

[Santiago]

I keep expecting some Apple instruction set extensions, specifically to accelerate x86-64 emulation in forthcoming Apple silicon.

They're present. Well, not quite instruction set extensions, but Apple Silicon chips will support switching between the memory models used by ARM and x86 chips in order to ease x86 emulation:

https://twitter.com/x/status/1288660944885821442

 

[Antony Newman]

The problem that NVIDIA have is that ARM v9 (and latter day ARM v8.n) allow ISA extensions as a first class citizen.

If RIKEN could already compete with NVIDIA in performance / watt without a GPU - and the next versions can directly accelerate 'GPU' and 'AI' loads - then NVIDIA would slowly be pushed out of the ARM super market.

As NVIDIA don't have an x86 license, and we are just 'around the corner' of Intel releasing an Xe GPU in a 3D stack (ie in the next 3 years), and AMD very likely to follow suit ... then NVIDIA would be squeezed out of that part of the market too.

This move would give NVIDIA early access to the next two process generations of TSMC chips, and allow ARM v9 to directly support ...... CUDA (by ISA extension) - especially as Nvidia have recently productionised the ARM CUDA supercomputer development support.

On the Networking side - this could also get very interesting : Mellanox + ARM's mesh + the fruits of Next Gen (Z) could see Nvidia being becoming an OEM supercomputer maker that pushes to disparate supers that could reduce their cooling bill dramatically.

If ARM was about encroach into NVIDIAs GPU territory - I wonder how the monopolies would view this?

AJ

 

[truthertech]

Apple already has an architecture license with ARM so I don’t see anything changing in the short term, but I could see this motivating Apple to develop their own instruction set in the long term.

Apple was a founder of ARM and used to own 20% of it before they sold off their share. They retain licensing rights and have no need or desire to purchase ARM.

 

[UnusedLoginID]

Why wouldn't Intel buy ARM?

Exactly! Intel is losing ground in the computer chipset, they lost the mobile devices chipset to Apple, buying ARM would bring the future of chips into their basket.
I’m also very concerned by NVIDIA acquiring ARM because they have not been the leaders at quality-made graphics chips. Just ask Apple and the numerous recalls/repair programs on 15” MacBook Pro.

 

[UnusedLoginID]

Arm is a competitor. It'd be tough to get regulatory approval.

Sprint was a T-mobile competitor. Didn’t prevent the acquisition.
There are many chip manufacturers out there, Intel buying ARM would go through, or NVIDIA would not be able to buy ARM.

 

[randian]

Apple already has a license agreement.

I don't think they've licensed ARMv9 yet.

 

[randian]

I wonder if future Apple Silicon is going to be ARM-less... when you control the whole stack, do you even need the ARM instruction-set anymore?

Apple still needs a compiler, so unless they're willing to contribute work on a custom instruction set to Clang I don't see them switching.

MIPS is my sentimental favorite. I don't know what it would cost to license. Probably pretty cheap though.

 

[Analog Kid]

This move would give NVIDIA early access to the next two process generations of TSMC chips

Correct me if I'm wrong, but I don't think ARM has a relationship with TSMC. ARM licensees can produce their products on TSMC if they choose to.

and allow ARM v9 to directly support ...... CUDA (by ISA extension) - especially as Nvidia have recently productionised the ARM CUDA supercomputer development support.

Couldn't Nvidia choose to do this anyway, even if ARM were a separate entity?

NVIDIA Brings CUDA to Arm, Enabling New Path to Exascale Supercomputing

International Supercomputing Conference -- NVIDIA today announced its support for Arm CPUs, providing the high performance computing industry a new path to build extremely energy-efficient, AI-enabled exascale supercomputers.

nvidianews.nvidia.com

• "Once stack optimization is complete, NVIDIA will accelerate all major CPU architectures, including x86, POWER and Arm."

Interesting that they spell it as "Arm" throughout the press release... "NVIDIA", "CUDA", "POWER", and "Arm". You think they're just negging the company so SoftBank will try harder to win their approval?

 

[Nugget]

Interesting that they spell it as "Arm" throughout the press release... "NVIDIA", "CUDA", "POWER", and "Arm". You think they're just negging the company so SoftBank will try harder to win their approval?

"Arm" is how it should be written.

"From 1 August 2017, Arm has a new look and feel. The business has a new corporate logo and it is now using the Arm word in sentence case instead of the ARM word in uppercase in text. We ask all of Arm’s customers, partners, licensees and any other third parties to use the Arm word in sentence case in text in all relevant materials. The only exception to this rule will be when using the ARM word in any circumstances, where all of the surrounding words also appear in uppercase, e.g. headings."

Trademarks – Arm®

This site contains a list of Arm’s trademarks, usage guidelines and a facility to request permission to use Arm logos and copyright material.

www.arm.com

 

[Analog Kid]

"Arm" is how it should be written.

"From 1 August 2017, Arm has a new look and feel. The business has a new corporate logo and it is now using the Arm word in sentence case instead of the ARM word in uppercase in text. We ask all of Arm’s customers, partners, licensees and any other third parties to use the Arm word in sentence case in text in all relevant materials. The only exception to this rule will be when using the ARM word in any circumstances, where all of the surrounding words also appear in uppercase, e.g. headings."

Trademarks – Arm®

This site contains a list of Arm’s trademarks, usage guidelines and a facility to request permission to use Arm logos and copyright material.

www.arm.com

Huh. Dated 2017... 😳

 

[Seoras]

I keep expecting some Apple instruction set extensions, specifically to accelerate x86-64 emulation in forthcoming Apple silicon.

Don't. The whole philosophy of RISC processor design is to keep the instruction set as simple as possible and move the complexities to the compiler.
Adding instructions just creates legacy in silicon which needs supporting in perpetuity. x86 was crippled by this.

 

[ZzapDK]

Why does Nvidia want an IP licensing business? Is this a trophy buy?

In order to fully move into the supercomputer marked for a start:
They own the GPUs, they own the networking and soon they will own ARM

https://blogs.nvidia.com/blog/2020/05/04/nvidia-acquires-cumulus/
https://www.top500.org/news/japan-captures-top500-crown-arm-powered-supercomputer/

 

[ZzapDK]

Nvidia are a very aggressive company. We were all wondering what Jenson had up his sleeve (he always has something, and it always turns out to be enough to save Nvidia for the next decade+).

I agree, their hubris is unequaled - anyone else but NVIDIA!
I cringed when I saw a Jensen Huang statue near the window in the RTX demo. Innocent, I know, but I couldn't feel but compare it to a cult idol

 

[Analog Kid]

In order to fully move into the supercomputer marked for a start:
They own the GPUs, they own the networking and soon they will own ARM

https://blogs.nvidia.com/blog/2020/05/04/nvidia-acquires-cumulus/
https://www.top500.org/news/japan-captures-top500-crown-arm-powered-supercomputer/

But why own it? Everyone else is fine licensing from them. ARM Arm is an IP company. I could imagine if someone saw an advantage in bundling IP. Nvidia is not an IP company. What's the advantage to Nvidia in owning Arm?

 

[Analog Kid]

Adding instructions just creates legacy in silicon which needs supporting in perpetuity. x86 was crippled by this.

Apple doesn't hesitate to jettison unnecessary features. x86 carries instructions into perpetuity as a business choice. I can imagine Apple deciding at some point that Big Sur will no longer run on processors made after 2030, for example-- or just trapping on the unimplemented instruction and handling it in software until applications are updated.

Use it for 5 years, deprecate it for 5 years, then either trap to a workaround or crash the outdated application.

 

[ZzapDK]

But why own it? Everyone else is fine licensing from them. ARM Arm is an IP company. I could imagine if someone saw an advantage in bundling IP. Nvidia is not an IP company. What's the advantage to Nvidia in owning Arm?

Yet there they are, discussing price

 

[JPack]

But why own it? Everyone else is fine licensing from them. ARM Arm is an IP company. I could imagine if someone saw an advantage in bundling IP. Nvidia is not an IP company. What's the advantage to Nvidia in owning Arm?

Intel to Pay NVIDIA Technology Licensing Fees of $1.5 Billion

SANTA CLARA, CA--NVIDIA announced today that it has signed a new six-year cross-licensing agreement with Intel. For the future use of NVIDIA's technology, Intel will pay NVIDIA an aggregate of...

nvidianews.nvidia.com

 

[Elijen]

Considering the apparent animosity Apple has toward NVIDIA, I wonder how comfortable Apple is going to be licensing from them, especially now with desktop processors. I wonder if future Apple Silicon is going to be ARM-less... when you control the whole stack, do you even need the ARM instruction-set anymore?

It's quite common for 2 companies to trade and sue each other simultaneously. Companies do not need to be friends to buy from each other. Just look at Apple vs Samsung.