Nvidia has approached Softbank and is considering buying ARM Holdings

[uzzi38]

Will update with articles as they come out (they haven't yet but news is public). For now just a couple of Tweets:

https://twitter.com/x/status/1286003982079533056

https://twitter.com/x/status/1286004990327132161

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

 

[dacostafilipe]

I don't see why he is wasting time and money on these alleged acquisition talks with SoftBank.

"Moving" the smartphone industrie to a Nvidia Cuda ecosystem sound like a really good investment, no?

Or, how about DL/ML stuff that somehow requires a specific amount of DGXs to really be usable?

You certainly lack imagination

 

[Vattila]

"Moving" the smartphone industrie to a Nvidia Cuda ecosystem sound like a really good investment, no?

Why would he need ARM to do that?

 

[dacostafilipe]

Why would he need ARM to do that?

Buying ARM is the same as buying the ecosystem.

 

[Kenmitch]

Buying ARM is the same as buying the ecosystem.

More like guiding the future of the ecosystem in the end.

 

[coercitiv]

More like guiding the future of the ecosystem in the end.

More like getting pole position in every arch refresh race, for free.

The best analogy I can think of for a successful acquisition is Nvidia would lend a very large sum today only to have their competitors pay interest forever.

 

[dacostafilipe]

More like getting pole position in every arch refresh race, for free.

I don't know, does Nvidia really care about the ARM arch for themselves?

They used ARM cores in the past, but they aren't they creating custom ones now?

I really think the benefit would be the "guiding the future of the ecosystem" as @Kenmitch said.

 

[Vattila]

Buying ARM is the same as buying the ecosystem.

The obvious problem with that is that they end up with an ecosystem of unwilling customers, if regulatory bodies around the globe should even tolerate it. There would be an accelerated migration to alternatives, primarily RISC-V, I guess (which many predict is the direction of open ISAs anyway).

The central philosophy of ARM's business model is that they do not compete with their customers, nor favour one over the other. They only compete on a technical level, by striving to produce the best IP and cores possible for licensing.

I presume Nvidia is not going to turn itself into a licensing house following this model. They are primarily a HPC product company, seeking to compete in high-margin growth markets. For these reasons, I don't see Nvidia taking over the licensing side of ARM.

PS. I was a shareholder of MIPS when they were broken up and the operational side was sold to Imagination. The control of the patent portfolio was taken over by a consortium to protect licensees. If ARM is sold/broken up, I guess it will be similar.

 

[NTMBK]

The obvious problem with that is that they end up with an ecosystem of unwilling customers, if regulatory bodies around the globe should even tolerate it. There would be an accelerated migration to alternatives, primarily RISC-V, I guess (which many predict is the direction of open ISAs anyway).

The central philosophy of ARM's business model is that they do not compete with their customers, nor favour one over the other. They only compete on a technical level, by striving to produce the best IP and cores possible for licensing.

I presume Nvidia is not going to turn itself into a licensing house following this model. They are primarily a HPC product company, seeking to compete in high-margin growth markets. For these reasons, I don't see Nvidia taking over the licensing side of ARM.

PS. I was a shareholder of MIPS when they were broken up and the operational side was sold to Imagination. The licensing side and control of the patent portfolio were taken over by a consortium to protect licensees. If ARM is sold/broken up, I guess it will be similar.

I don't think such a breakup could work. ARM's "licensing" business is mostly around licensing IP to customers- if NVidia is buying ARM to get control of the CPU design teams, then what is there for a spun-off licensing company to license out? A bunch of rapidly aging designs that won't see further updates, and aren't optimized for new manufacturing processes?

 

[dacostafilipe]

...

A leader in ML/DL hardware that wants to add IP to the ARM portfolio to allow small companies to easier integrate into their workflow or not having the need to create their own hardware. I don't think it sounds that bad. cuDNN all over right? (cuda lockdown included )

And if Nvidia would add Cuda to all future ARM SoC, they will certainly also stop Xavier and co, so no competition.

 

[Vattila]

if NVidia is buying ARM to get control of the CPU design teams, then what is there for a spun-off licensing company to license out?

Good point. In the MIPS case, the consortium basically controls the patent portfolio to protect licensees. It is made up of many stakeholders in the MIPS architecture. I'm not sure they do any meaningful new licensing revenue. But the protection part of the consortium obviously was important enough for many competing companies using MIPS to get together. On the licensing side, Imagination is itself a licensing business, so they acquired MIPS for the purpose of adding CPU capability to their IP portfolio available for licensing, so that they could be more of an equal competitor against ARM (which had acquired the Mali designers, adding GPU capability). In many ways it is very similar to how Nvidia is seeking CPU capability today, but at the same time very different, since Nvidia is not primarily a licensing house. Nvidia just wants to boot EPYC out of their DGX A100 server, and replace it with their own CPU and system platform.

 

[DrMrLordX]

There would be an accelerated migration to alternatives, primarily RISC-V, I guess (which many predict is the direction of open ISAs anyway).

I don't think RISC-V can be taken seriously as a competitor to ARM's larger cores. It's too squarely-aimed at embedded microcontroller markets and the like. Cortex-M? Sure, it'll compete with that. Matterhorn? Nah.

 

[Vattila]

I don't think RISC-V can be taken seriously as a competitor to ARM's larger cores. It's too squarely-aimed at embedded microcontroller markets and the like. Cortex-M? Sure, it'll compete with that. Matterhorn? Nah.

Interestingly, Nvidia is already a big promoter and adopter of RISC-V:

"Along with other major industry players Google, Samsung, IBM, and Qualcomm, NVIDIA is a member of the RISC-V Foundation, and has long supported RISC-V development. In 2016 the company unveiled plans to replace the internal micro-controllers of their graphic cards with next-gen RISC-V-based controllers built for upcoming NVIDIA GPUs."

Is NVIDIA Doubling Down On RISC-V?

The US chip giant now appears to be exploring the terrain beyond GPU architecture, raising the stakes on other dedicated electronic circuits

medium.com

But, as you point out, this is mainly for embedded security processors and controllers. I guess they see the road to competing in HPC with RISC-V too long and winding, hence their interest in buying ARM's team and roadmap instead, giving them an opportunity to jump into leadership position in the fledging ARM server market segment.

 

[Thibsie]

I don't think RISC-V can be taken seriously as a competitor to ARM's larger cores. It's too squarely-aimed at embedded microcontroller markets and the like. Cortex-M? Sure, it'll compete with that. Matterhorn? Nah.

Not right now. In 5 years?

 

[Doug S]

I don't think RISC-V can be taken seriously as a competitor to ARM's larger cores. It's too squarely-aimed at embedded microcontroller markets and the like. Cortex-M? Sure, it'll compete with that. Matterhorn? Nah.

The exact same statements were made regarding ARM in the recent past. There's nothing preventing someone from designing a RISC-V core that's competitive on the high end, it just hasn't happened yet. Just like there was nothing preventing someone from designing an ARM core that was competitive on the high end, until it did happen.

 

[Thala]

The exact same statements were made regarding ARM in the recent past. There's nothing preventing someone from designing a RISC-V core that's competitive on the high end, it just hasn't happened yet. Just like there was nothing preventing someone from designing an ARM core that was competitive on the high end, until it did happen.

It was only possible because the ARM ecosystem was ready for it. From this perspective RISC-V cant be taken seriously - there is simply no driver for it except some embedded use-cases. Things do not just happen, you know? There is not even any value proposition of RISC-V outside of embedded.

 

[Thibsie]

It was only possible because the ARM ecosystem was ready for it. From this perspective RISC-V cant be taken seriously - there is simply no driver for it except some embedded use-cases. Things do not just happen, you know? There is not even any value proposition of RISC-V outside of embedded.

NVIDIA might just make it happen.

 

[Doug S]

It was only possible because the ARM ecosystem was ready for it. From this perspective RISC-V cant be taken seriously - there is simply no driver for it except some embedded use-cases. Things do not just happen, you know? There is not even any value proposition of RISC-V outside of embedded.

The driver for it would be what people are worried about here - that Nvidia buys ARM and makes life hard for some of the current ARM licensees by raising prices, withholding the fastest designs for their own use, or whatever.

China is another potential driver of this, given the ever escalating trade tensions I could see the state deciding they want to put their effort behind RISC-V rather than ARM to guarantee no future entanglements.

The ecosystem is ready for it when there is market demand for it. That could be coming through either of those paths, depending on if ARM is sold and to whom, and what happens with the piece of ARM that's a JV in China.

 

[NostaSeronx]

It's too squarely-aimed at embedded microcontroller markets and the like.

The more simple ISA has benefits in those markets, but the cores for RISC-V have no such targets.

A RISC-V core can be anything; HPC core, GPU core, Wide CPU core, Thin CPU core, controllers, DSPs, etc.

RISC-V in the prototype cores are in fact squarely-aimed at the Green500 which is HPC and the like.

 

[A///]

When Nuvia has a physical product to showcase then we can discuss the pros and cons of RISC-V vs. ARM. I'm surprised their little benchmark article stunt from earlier this month wasn't called out more than it was.

 

[DrMrLordX]

Not right now. In 5 years?

The exact same statements were made regarding ARM in the recent past. There's nothing preventing someone from designing a RISC-V core that's competitive on the high end, it just hasn't happened yet. Just like there was nothing preventing someone from designing an ARM core that was competitive on the high end, until it did happen.

I'll address both of these points as best I can, since there seems to be a lot of enthusiasm about RISC-V out there in circles that don't (necessarily) like to discuss what RISC-V actually is. Actually, articles like this one can express it better than I:

RISC-V Bases and Extensions Explained - CNX Software

RISC-V ISA nonemclature is comprised of RISC-V bases and RISC-V extensions. In this post, we explains how to decode RISC-V core architecture names.

www.cnx-software.com

No two RISC-V SoCs need be alike. And let's not forget the vendor-specific extensions that may wind up being proprietary. If I were to write software for RISC-V, unless I knew my exact target SoC in advance, it would be impossible to know exactly which extensions would and would not be supported. That's fine for embedded since, yeah, you know your hardware intimately in advance. It might even be okay for consoles. But for desktop/laptop consumer software? ISA extension hell, here we come! And you thought AVX512 was bad.

 

[Doug S]

No two RISC-V SoCs need be alike. And let's not forget the vendor-specific extensions that may wind up being proprietary. If I were to write software for RISC-V, unless I knew my exact target SoC in advance, it would be impossible to know exactly which extensions would and would not be supported. That's fine for embedded since, yeah, you know your hardware intimately in advance. It might even be okay for consoles. But for desktop/laptop consumer software? ISA extension hell, here we come! And you thought AVX512 was bad.

Who's talking about desktop/laptop? ARM is barely there, it isn't worth worrying about in the context of RISC-V potentially replacing ARM in some markets.

The potential issue if Nvidia buys ARM and tightens the screws is embedded and Android. There isn't a laptop/desktop market for it to lose so there wouldn't be any need to go RISC-V there.

For Android, either Qualcomm's implementation would become the defacto ISA standard, or Google would be forced to step in and set some ground rules. Though no doubt companies like Samsung would take the opportunity to add some extensions they would claim improves performance, but only their own apps would use them.

 

[A///]

How do you figure Google can make a play here? Ban QC SoC from running Android at a hardware check level?

 

[Thala]

The driver for it would be what people are worried about here - that Nvidia buys ARM and makes life hard for some of the current ARM licensees by raising prices, withholding the fastest designs for their own use, or whatever.

China is another potential driver of this, given the ever escalating trade tensions I could see the state deciding they want to put their effort behind RISC-V rather than ARM to guarantee no future entanglements.

The ecosystem is ready for it when there is market demand for it. That could be coming through either of those paths, depending on if ARM is sold and to whom, and what happens with the piece of ARM that's a JV in China.

Sure, if the market situation changes - i was talking about the status quo, ´which is nor particularly favorable for RISC-V in markets outside embedded.

 

[NostaSeronx]

But for desktop/laptop consumer software? ISA extension hell, here we come! And you thought AVX512 was bad.

If there was a desktop/laptop-orientated RISC-V solution that was successful. Then, all competing solutions will support all the extensions that the successful solution had.

Much like the proposed x86 feature levels:
Base, Level A, Level B, Level C, Level D.

RISC-V can do the same:
Base, Level A, Level B, Level C, Level D

Base = bare minimum to support

There will be a lot less extension rot in RISC-V in comparison. Since, most of the levels for x86 are 128-bit, 256-bit, 512-bit extensions. Meanwhile, RVV in real-use so far extends beyond that: XT-910 can support the operation width from 64 bits to 1024 bits.

 

[DrMrLordX]

Who's talking about desktop/laptop?

That's where ARM is headed, if it can get traction. And it's headed for the server room, too. Or at least it's trying. Those are the markets where NV needs ARM the most - not RISC-V. NV is not looking at buying ARM to take over the smartphone market.