What if Nvidia had acquired Arm?

Jay Goldberg

Posts: 74   +1
Staff
Editor's take: Nvidia's is now hailed for its single-minded vision of giving birth to AI. But the real lesson from their success is better demonstrated by their failures. This is a company that is not afraid of taking risks.

Back in business school, I had secured an interview for a really great job. The interview was scheduled early in the morning, but I woke up early. Had some coffee. Reviewed my notes. I was ready. This was long enough ago that I was using a cordless, landline phone to call in.

I picked up the phone and there was no dial tone. I burst into my roommate's room, but his wired phone did not work either. I then turned to my cell phone, but I lived in a large dark zone in the coverage map and could not connect that way either. A complete telecoms failure.

Editor's Note:
Guest author Jonathan Goldberg is the founder of D2D Advisory, a multi-functional consulting firm. Jonathan has developed growth strategies and alliances for companies in the mobile, networking, gaming, and software industries.

I then had to run around the neighborhood until I found a payphone at a gas station a few blocks from the apartment. (Yes – cordless, landlines, dial tones, urban coverage holes, and payphones – this was a long time ago.) Standing in the middle of a gas station at a busy intersection I finally connected, 30 minutes late. The managing director's assistant answered the phone and told me "Yes, he understands you were delayed, and asks that you call back when you actually want this job."

I was prompted to think about this recently while listening to sessions at Nvidia's GTC developer conference. At one point, someone asked CEO Jensen Huang about their Hopper CPU in its relation to other Arm-based CPUs, and it occurred to me that not so long ago, Nvidia tried really hard to buy Arm. Where would Nvidia be today if that deal had happened?

Nvidia abandoned the deal in early 2022. If the deal had gone through it likely would not have closed until the end of that year. And of course, late that year in November, OpenAI launched ChatGPT and turned the world upside down.

Given that timing, If the deal had gone through, it is very possible that the integration of Arm would have become a second tier problem. Nvidia would still be riding high, and it is hard to see the management team devoting as much energy to boring integration issues when they could literally become AI rockstars. Arm probably would have suffered as a result, still large, but languishing as a bit of an afterthought.

By comparison, what if the deal had sailed through and closed earlier in 2022 and become a focus of Nvidia. Would they have missed ChatGPT? Probably not, Nvidia had been betting heavily on AI for a long time and it is unlikely they would have let the opportunity slip, but they may not have executed as well as they are now, constantly dealing with Arm integration and trying to shoehorn that narrative into the AI narrative.

We mention this because Nvidia's GTC has become a big event – celebrating Nvidia's rise to the top of the semiconductor ranks. They took some big risks and had the foresight to invest heavily in machine learning, AI, transformers and all the rest of what they now label "Accelerated Compute." The company, and their CEO in particular, deserve immense credit for sticking to their vision.

This is not the story of a company with a single focus, sticking to their guns amidst a sea of naysayers. Instead, we see this as a company that has never been afraid to take chances...

That being said, I think that's the wrong lesson to take away from Nvidia's success. In hindsight, it is easy to say that Nvidia saw the future and reshaped the company to become the leader in AI, but that is not what happened. This is not the story of a company with a single focus, sticking to their guns amidst a sea of naysayers. Instead, we see this as a company that has never been afraid to take chances, to try out many different things, and accept that some of them will end in failure.

Dig through the archives and you can find a lot of things the company did not do right. Not so long ago, they were the leader in silicon for crypto digital coins. At GTC, Huang said the company was probably out of that market entirely. For years, all they talked about were autonomous vehicles. Those are still somewhere in the mix, but garner very little attention nowadays.

Anybody remember 'Icera' who Nvidia acquired a decade ago to make a stab at competing with Qualcomm for cellular modems? No worries, no one else remembers that either.

Arm could very well have ended up as another one of those diversions. We never thought the strategic case for that deal was terribly solid. They could have made it work, but it always risked being a distraction. Again, in hindsight, this is another one of those deals which ended up better for it never happening.

So as you listen to Huang's keynotes, and hear all the talk about their vision of the future – AI factories, Omniverse, 6G (lord, help us), and all the rest – take them all with a grain of salt. They are good ideas (mostly). Some of them could turn out to be incredible businesses. But they each matter less than the fact that Nvidia is willing to try all of them out.

The company's vision of the future, like its products, is not deterministic. Nvidia is playing, and playing, the odds, and that ability to take risks is the key to their success.

Permalink to story:

 
From my understanding, NVIDIA move on ARM, was about the licensing deals and patents. Nvidia with that "power" could have made licensing deals with third partys adding Nvidia GPU's IP into ARM ecosystem instead of current GPU options in ARM ecosystem, they tried to make Tegra SoC, but they lack the ARM know-how on CPU side of things, and Nvidia doesnt like to pay for "licensing" neither giving development to other companys that can sell their inputs.... they have a thirst for control back-to-back on tech developments.

Nvidia have a long history of intents of grabbing a CPU branch with the right Patents/Licensing, to make a colossal impact on Datacenter and Server infrastrutures. Blackwell is a half-baked vision of the SoC Nvidia's wants to bring to life.

I can imagine, Apple Arm CPU's, with NVIDIA developments on it. Nvidia Engineers are really great at making things work better and faster. Apple would pay top dollars for that kind of engineering.
 
IF that had happened, you would see vast investment into RISC-V right now.
From my understanding, NVIDIA move on ARM, was about the licensing deals and patents. Nvidia with that "power" could have made licensing deals with third partys adding Nvidia GPU's IP into ARM ecosystem instead of current GPU options in ARM ecosystem, they tried to make Tegra SoC, but they lack the ARM know-how on CPU side of things, and Nvidia doesnt like to pay for "licensing" neither giving development to other companys that can sell their inputs.... they have a thirst for control back-to-back on tech developments.

Nvidia have a long history of intents of grabbing a CPU branch with the right Patents/Licensing, to make a colossal impact on Datacenter and Server infrastrutures. Blackwell is a half-baked vision of the SoC Nvidia's wants to bring to life.

I can imagine, Apple Arm CPU's, with NVIDIA developments on it. Nvidia Engineers are really great at making things work better and faster. Apple would pay top dollars for that kind of engineering.
Nvidia tried, multiple times, to make ARM CPUs with the Tegras. They were never competitive with snapdragons, and fell further behind as time went on. I'm pretty sure apple would be better off not needing to use Nvidia designs.....
 
Nvidia has turbocharged the ARM core but could not market it due to the ARM licensing clauses. They would essentially have lost that IP to all the permanent residents of the ARM ecosystem (you know most of them), without a way to recoup the development cost nor profit from the market advantage.
Nvidia has lots of completely different core types, the ARM one is obviously well liked. Nvidia will still turbocharge ARM designs despite others, its good to have competition, if they can keep up.
ARM and RISC are too fat to be good AI inference processors, hopefully Nvidia can chunk something much smaller.
 
I can forsee Arm still being in consumer products for many years to come but enterprise and web infrastructure could very well possibily move to RISC-V and maybe that's where Nvidia is better positioning it's CPU side.
 
I actually think its a shame Nvidia hasnt acquired ARM. They are a very successful and innovative company that clearly has a good culture. It would have been interesting to see what they would have done with it.
 
IF that had happened, you would see vast investment into RISC-V right now.
Nvidia tried, multiple times, to make ARM CPUs with the Tegras. They were never competitive with snapdragons, and fell further behind as time went on. I'm pretty sure apple would be better off not needing to use Nvidia designs.....

I guess thats why Nintendo went Tegra for Switch and will do the same for Switch 2? :laughing:

Actually it was Nintendo that wanted Tegra badly.

Nintendo want Tegra again for Switch 2 - mostly for DLSS which beats any other upscaling method.
 
I guess thats why Nintendo went Tegra for Switch and will do the same for Switch 2? :laughing:

Actually it was Nintendo that wanted Tegra badly.

Nintendo want Tegra again for Switch 2 - mostly for DLSS which beats any other upscaling method.
Nintendo wasted hundreds of millions of dollars if not billions into the WiiU. They just got an offer the shelf product that Nvidia had already been shipping for over a year before Nintendo decided they were going to use it in their design as they needed to get a product out of the door to replace the 3DS (their only money maker at the time) and the loss maker the WiiU.

The Switch is a flawed product in many ways but somehow it just works and the low level API from Nvidia and the lightness of the system OS means that the Switch has done surprisingly well in performance for such a low powered device. The Switch 2 will have been designed a lot better as Nintendo will have the funds and time to do so this time although I do still feel it'll be slightly off the shelf of an existing Nvidia Tegra product.

I was looking at the Tegra wiki page and I believe the likelihood is the Switch model will have 1024 CUDA cores setup and the CPU cores at 1.5Ghz with 8GB of LPDDR5X memory producing 68GB/s memory bandwidth in portable and 102.4GB/s in docked with around compute of 1.28tf in portable and 1.88tf in docked (FP32) as this configuration would be between 7-15w of TDP. So any rumour reports of more CUDA cores etc I believe is incorrect.
 
Nintendo wasted hundreds of millions of dollars if not billions into the WiiU. They just got an offer the shelf product that Nvidia had already been shipping for over a year before Nintendo decided they were going to use it in their design as they needed to get a product out of the door to replace the 3DS (their only money maker at the time) and the loss maker the WiiU.

The Switch is a flawed product in many ways but somehow it just works and the low level API from Nvidia and the lightness of the system OS means that the Switch has done surprisingly well in performance for such a low powered device. The Switch 2 will have been designed a lot better as Nintendo will have the funds and time to do so this time although I do still feel it'll be slightly off the shelf of an existing Nvidia Tegra product.

I was looking at the Tegra wiki page and I believe the likelihood is the Switch model will have 1024 CUDA cores setup and the CPU cores at 1.5Ghz with 8GB of LPDDR5X memory producing 68GB/s memory bandwidth in portable and 102.4GB/s in docked with around compute of 1.28tf in portable and 1.88tf in docked (FP32) as this configuration would be between 7-15w of TDP. So any rumour reports of more CUDA cores etc I believe is incorrect.
The magic of Switch 2 will be full backwards comp. + DLSS/DLAA.
It should be able to run pretty much all Switch games at locked 60 fps in both portable and docked mode and I hope thats the goal for new games as well. Or atleast give the user the choice between 30 and 60.
 
Back