News Report: Nvidia's Getting Serious About Acquiring Arm

[Admin]

A new report from Bloomberg claims Nvidia is the only company in the run to buy Arm.

Report: Nvidia's Getting Serious About Acquiring Arm : Read more

 

[Chung Leong]

A big rationale behind the deal, I think, is getting Nvidia's AI tech into ARM-powered devices. That helps solidify CUDA's position as the platform of choice. Tensor cores in phones can also help Nvidia's cloud gaming effort. Using DLSS on the receiving end could greatly reduce the bandwidth necessary.

 

[spentshells]

I like this idea, I was always surprised Apple used AMD products to begin with, Just imagine the price of a MacBook pro with a Nvidia workstation graphics card. Oooof

 

[castl3bravo]

Eewwwww. Nvidia's antics aren't a good sign of things to come if the acquire Arm successfully. Arm's intellectual property extends to "smart" devices everywhere. Not just Apple, but cellphones, WiFi routers, Microwaves, refrigerators, doorbells. Oh and what about 5G cellphone networks? These networks will be a heavy user of Arm based CPUs for 5G enabled "Internet of Things" (IoT) devices you'll be buying a couple of years. I predict an Nvidia tax for all those new 5G enabled smart devices.

 

[bit_user]

A big rationale behind the deal, I think, is getting Nvidia's AI tech into ARM-powered devices.

Sorry, I don't agree with this, at all. ARM already has its own AI cores. And last time Nvidia tried to play in the phone market, it didn't go very well.

Arm Announces Ethos-N78 NPU: Bigger And More Efficient

www.anandtech.com

That helps solidify CUDA's position as the platform of choice.

No, it doesn't. Google doesn't want Android apps using CUDA or any other APIs besides the ones officially included in the Android spec, and there's no way Google is adding CUDA to Android.

Android already abstracts the underlying technology used in deep learning acceleration, so you're insulated from whatever lower-level technology the device is using.

BTW, Nvidia already partnered with ARM to integrate its deep learning tech in their platform:

Under this partnership, NVIDIA and Arm will integrate the open-source NVIDIA Deep Learning Accelerator (NVDLA) architecture into Arm’s Project Trillium platform for machine learning. The collaboration will make it simple for IoT chip companies to integrate AI into their designs and help put intelligent, affordable products into the hands of billions of consumers worldwide.

https://nvidianews.nvidia.com/news/...ring-deep-learning-to-billions-of-iot-devices​

​

I think Nvidia did that because it wanted to create downstream demand for its training GPUs and infrastructure. You don't open-source something like NVDLA if you actually want to monetize that market.

 

[bit_user]

I like this idea, I was always surprised Apple used AMD products to begin with, Just imagine the price of a MacBook pro with a Nvidia workstation graphics card. Oooof

Given that Apple is now also making its own GPUs for its iPhones, it could be a matter of time before they just scale these up for their laptops, like they're doing with the CPUs.

 

[bit_user]

I predict an Nvidia tax for all those new 5G enabled smart devices.

This could just hasten the industry's transition to RISC V.

 

[Chung Leong]

Sorry, I don't agree with this, at all. ARM already has its own AI cores. And last time Nvidia tried to play in the phone market, it didn't go very well.

If you can't beat them, buy them out 🙂

 

[thisisaname]

One thing is for certain they would be stupid to pay anywhere near $32 billion for them.

Placing this licensor in the hands of Nvidia would give the graphics card maker a lot of power in the market.

They flex that power to much RISC V would get a lot more support.
ARM's business model is make something good that people want them licence it relatively cheaply. They make a nice amount from licensing from volume of devices not a big up front cost.

In four years under SoftBank's direction, Arm's annual revenue has only increased from $1.2 billion to $1.9 billion, as reported by the Financial Times. Under a new roof, such as Nvidia's, there is a chance that Arm could meet higher expectations. Nvidia's revenue has tripled in that same timeframe.

Total different business you squeeze licensees to much and that is going to kill the business.

 

[bit_user]

One thing is for certain they would be stupid to pay anywhere near $32 billion for them.

I agree with this, but mainly because of China looking to move away from Western tech and the threat RISC V now poses. Both of those factors significantly limit ARM's growth potential.

In the near term, I think the biggest effect of RISC V will be simply to exert downward pricing pressure on ARM. From the time they became a dominant player until now, they could basically charge as much as the market would bear.

 

[jtremblant]

For those happy for this allegedly new Nvidia adquisition, think again you will be very likely paying twice for an ARM IP license and that will reflect upon not only ARM customers but end users as well. There's only one thing Nvidia excels at, making money by overcharging for their products. Also, consider the return of investment for the green company regarding this huge expense. Apple will be moving in 5 years to RISC-V once Nvidia start hiking up ARM IP license fees every year.

 

[bit_user]

you will be very likely paying twice for an ARM IP license and that will reflect upon not only ARM customers but end users as well.

That's certainly towards the negative end of the plausible outcomes.

I hope this doesn't go through. But, if it does, I expect Nvidia will take a reasonably long-term view towards the pricing model. They can't realistically recoup their investment in a short or medium time frame, so they need to figure out how to keep the ARM business model going into the long term. And I believe they're not too dense to see that overcharging for its IP runs counter to that.

My sense is that their interest comes from wanting early access to the latest server & embedded ARM CPU core IP, as well as perhaps ensuring that it plays well with all of their other products and technologies. They've got to be looking at this, strategically.

There's only one thing Nvidia excels at, making money by overcharging for their products.

The only way that works is if your market position is sufficiently strong. So, companies like Apple and Nvidia first need to focus on excellence. If we're being fair, I think they've achieved it. Both have their loyal fans (Apple more so than Nvidia), but that good will would dwindle if they failed to keep delivering on expectations.

Apple will be moving in 5 years to RISC-V once Nvidia start hiking up ARM IP license fees every year.

It could happen. Right now, Apple and ARM need each other. However, by the time Apple clears its ARM transition, we could be looking at RISC VI or something else.

 

[bit_user]

I'm not sure how many know this, but AMD actually launched a small line of ARM server CPUs, a couple years before Ryzen:

https://www.amd.com/en/amd-opteron-a1100

Those just use ARM's Cortex-A57 cores, but Jim Keller was said to have designed a fully-custom ARM core, before he left AMD.

https://en.wikipedia.org/wiki/AMD_K12​

Rumor has it, that was mothballed since AMD lacked the resources to bring both it and Zen to market. Given the nascent state of the ARM server market, at the time, it only made sense that K12 should get the short end of the stick.

Well, if this acquisition goes through, then I predict AMD will leave ARM firmly in its rearview, and look elsewhere for any post-x86 plans.

 

[Chung Leong]

I hope this doesn't go through. But, if it does, I expect Nvidia will take a reasonably long-term view towards the pricing model. They can't realistically recoup their investment in a short or medium time frame, so they need to figure out how to keep the ARM business model going into the long term. And I believe they're not too dense to see that overcharging for its IP runs counter to that.

ARM is reportedly considering massive licensing fee increases. An acquisition by Nvidia is probably the only thing that'd forestall them. As an independent company, ARM simply hasn't got many ways to generate growth.

 

[bit_user]

As an independent company, ARM simply hasn't got many ways to generate growth.

Well, they can expand the amount of IP they sell to existing customers, like they've done with their Mali GPUs and Ethos NPU blocks. They've also got interconnect technologies, but I don't think they've done a low-power ISP/DSP block. They're also expanding into new markets, like IoT, HPC, and cloud.

Really, until Trump ushered in a round of controls around partnerships and collaborations (which Japan later copied, with export bans on critical industrial chemicals needed by Samsungs fabs - this is significant, as the Japanese Softbank owns ARM), it looked like ARM's rise was effectively unbounded.

I think it's only the sense of unease created by those recent events, along with the up-and-coming alternative ISA of RISC V that has its investors contemplating a liquidation of ARM. Still, I think it's too soon to write them off.

 

[Deicidium369]

A big rationale behind the deal, I think, is getting Nvidia's AI tech into ARM-powered devices. That helps solidify CUDA's position as the platform of choice. Tensor cores in phones can also help Nvidia's cloud gaming effort. Using DLSS on the receiving end could greatly reduce the bandwidth necessary.

They can do that with a license, and does not require them buying the whole company. Licensing (which is 100% of ARMs income) would likely take a huge hit - since not everyone wants Nvidia's IP in their designs.

Tensor cores are Google - Google invented them, wrote the software framework for them, and would likely have an issue with Nvidia adding tensor cores into an IP/Designs they are selling. So Nvidia's AI Tech is actually Google AI Tech.

DLSS is software - leverages the Google Tensor cores.

CUDA is as solidified as solidified can get - it's a virtual walled garden - almost impenetrable.

Nvidia or Apple buying ARM is a terrible idea - once it ceases to be vendor agnostic, it loses what makes it valuable and successful.

 

[Deicidium369]

Well, they can expand the amount of IP they sell to existing customers, like they've done with their Mali GPUs and Ethos NPU blocks. They've also got interconnect technologies, but I don't think they've done a low-power ISP/DSP block. They're also expanding into new markets, like IoT, HPC, and cloud.

Really, until Trump ushered in a round of controls around partnerships and collaborations (which Japan later copied, with export bans on critical industrial chemicals needed by Samsungs fabs - this is significant, as the Japanese Softbank owns ARM), it looked like ARM's rise was effectively unbounded.

I think it's only the sense of unease created by those recent events, along with the up-and-coming alternative ISA of RISC V that has its investors contemplating a liquidation of ARM. Still, I think it's too soon to write them off.

What is it with you and Trump - they guy has no intelligence, zero business accumen and even less of strategy. The Apprentice was 100% scripted - he made a small fortune by starting with a large one. He is absolutely ignorant of the world around him (Who would have though that health insurance would be so complicated or Juneteenth - which he found out about hours earlier - declared that he is making it matter - also the same moron that met with N Korea - a meeting with the US or another World Power was what they have been seeking since the armistice - and Trump thinks it was because of him they wanted to meet - and exchange love letters - he gets played in every possible way - Rocketman gets a hugely beneficial photo op (several actually) and on top of that gets the US S Korea Naval exercises to cease - while we get NOTHING.) Trump did nothing - and anything that did happen was in no way shape or plan of his. Like once in a while a parrot seems to make a cogent statement in context to the actual conversation - and you consider for a second that - man the parrot makes a good point - then the parrot goes on about the great men of the confederacy.

Kodak is making a play into technical and specialty chemicals (coming full circle in essence)... Japan doesn't need Trump's help or example to want to screw over anything Korean - as the Japanese consider the Koreans to be subhuman.

Those targeted export controls over certain tech and certain processes have been part of the world for decades - at one point was more aimed at the Soviet Union/Russia than at China. Again, nothing to do with Trump.

RISC V is a novelty - 10 years from now it will still be a novelty, RISC V faces the same headwinds that ARM faces trying to go desktop - the MASSIVE library of code written for x86. Add the term Open Source to anything and the kiddie start cheering - having zero clue what it actually means.

 

[Chung Leong]

Well, they can expand the amount of IP they sell to existing customers, like they've done with their Mali GPUs and Ethos NPU blocks. They've also got interconnect technologies, but I don't think they've done a low-power ISP/DSP block. They're also expanding into new markets, like IoT, HPC, and cloud.

It takes time to develop new IP. Meanwhile, ARM's crown jewel patents are getting closer to expiration. Once the protection is gone, it'll be much easier for competitors to design processors as energy-efficient as ARM's offering, be it RISC-V or x86. Now is the time to sell the company...or cosplay the Reapers from Mass Effect.

 

[bit_user]

Tensor cores are Google - Google invented them, wrote the software framework for them, and would likely have an issue with Nvidia adding tensor cores into an IP/Designs they are selling. So Nvidia's AI Tech is actually Google AI Tech.

No. Just no. Google's TPU was nothing like what Nvidia calls it's "Tensor Cores". Either stick to what you actually know, or at least qualify your statements appropriately.

You can read about each, here:

​

https://www.anandtech.com/show/1174...erformance-analysis-live-blog-3pm-pt-10pm-utc​

https://www.anandtech.com/show/12673/titan-v-deep-learning-deep-dive/3​

​

Nvidia or Apple buying ARM is a terrible idea - once it ceases to be vendor agnostic, it loses what makes it valuable and successful.

Agreed. With Nvidia selling ARM-based SoCs and maybe more, that would tilt the market in its favor, souring many others on the platform, entirely.

 

[bit_user]

What is it with you and Trump

I'm not trying to make this about him, but rather I'm pointing out a consequence of this virtually unprecedented thing he did, which was to intervene in a major tech business arrangements. What a lot of people don't understand about a deal like Huawei licensing a core from ARM is that it's not just a simple transaction. There are services, support, and multiple deliverables included in the deal, which are essential for the customer's success and span virtually the life of the product using the part or IP. So, it's a long-term business arrangement and the potential for it to get seized up can only be regarded as an existential risk of the sort businesses absolutely hate.

The effects of that one move will be felt for a long time, to come. It can only be seen as a nail in the coffin of the US tech industry, though not even limited just to tech. It really needs to be struck down in a court case, to restore any degree of faith in US-based suppliers to fulfill contracts with international partners and customers.

The Apprentice was 100% scripted

I don't think quite 100%, however they did do multiple takes and used editing to show Trump in the best light, since the entire premise of the show hinged on the belief that he was some kind of business genius.

You make other interesting points, which I'm leaving out so as not to derail the discussion.

Japan doesn't need Trump's help or example to want to screw over anything Korean

I'm aware of the long-standing animosity and reasons behind it. However, I think the recent move to ban photoresist ingredients was unprecedented and possibly inspired by Trump's bans.

Supply of DRAM, NAND & Displays Could Be Disrupted by Japan & South Korea Dispute

www.anandtech.com

Those targeted export controls over certain tech and certain processes have been part of the world for decades - at one point was more aimed at the Soviet Union/Russia than at China. Again, nothing to do with Trump.

That was pre-globalization and probably didn't have the same level of disruption. What concerns me about Trump's move is that it came without warning and is the first instance I'm aware of, in the modern tech economy, of interfering with business relationships on this scale.

RISC V is a novelty - 10 years from now it will still be a novelty,

I don't agree, but we can agree to disagree. It's genuinely speculative, and I respect your opinion.

 

[Chung Leong]

Agreed. With Nvidia selling ARM-based SoCs and maybe more, that would tilt the market in its favor, souring many others on the platform, entirely.

Nvidia sells technologies at the end of the day. Is there really a fundamental difference between selling chip designs and actual chips? The only difference is who places the order at TSMC. I can easily see Nvidia adopting the ARM business model with their consumer GPUs. Instead of selling chips, the company would license their designs to video-card makers. That'd kill multiple birds at once. It offloads the risk of holding inventory to the card makers. At the same time, it resolves the cannibalization issue concerning their cloud gaming effort. Finally, it totally screws AMD since vendors holding a Nvidia license probably won't offer alternatives that competes with "their own chips."

 

[bit_user]

Nvidia sells technologies at the end of the day.

Jensen recently referred to it as primarily a software company.

Is there really a fundamental difference between selling chip designs and actual chips?

Yes, of course there is. Without getting into the nitty gritty details, consider why Apple pulled away from its position of officially supporting Mac clones, some 2 decades ago. The quality and nature of their software is what allows them to charge so much for the hardware. Granted, Apple hardware is typically nice, but not so nice as to justify the price differential vs. comparable PC products.

Instead of selling chips, the company would license their designs to video-card makers.

Most of those companies probably lack the resources, expertise, or desire to be in the chip business. GPUs are very cutting-edge, as well, which makes them more costly and risky.

That'd kill multiple birds at once. It offloads the risk of holding inventory to the card makers. At the same time, it resolves the cannibalization issue concerning their cloud gaming effort.

Look at how many ARM cores are shipped, annually, vs. ARM's revenues. It seems you can't charge nearly as much for the raw IP.

Finally, it totally screws AMD since vendors holding a Nvidia license probably won't offer alternatives that competes with "their own chips."

A significant number of video card brands are already 2nd party, like EVGA and Sapphire.

Plus, there's nothing to say that someone wouldn't license NVidia's IP and fab it into a chip that it sells on to whichever video card brands want to buy them. In fact, this is what would probably happen, due to the above points about the risk, costs, and complexity of getting a chip fabbed, packaged, and debugged. The net effect is that by adding another intermediary in the chain, it would squeeze Nvidia's margins.

 

[Chung Leong]

Most of those companies probably lack the resources, expertise, or desire to be in the chip business. GPUs are very cutting-edge, as well, which makes them more costly and risky.

They would, of course, hire Nvidia to do all that work. The only thing that changes is who makes the decisions.