Nvidia CEO Jensen Huang says ARM’s been too specific, needs to be a broad computing platform
Nvidia today held a virtual version of its GTC developer conference, and CEO Jensen Huang took questions from reporters in the afternoon.
Asked by ZDNet how Nvidia will do better than ARM, which it is buying for $40 billion, Huang said ARM's chip technology has been too narrow, despite enormous success.
"ARM sold 22 billion chips last year, it's a shocking number," said Huang, referring to the number of processors of all kinds in 2019 that licensed ARM's CPU designs. "Ninety-nine of that are system-on-chip," or SOC, he said, referring to "bespoke" chips that are designed for a variety of applications, everything from mobile phones to utility meters to smart lighting.
Also: It's official: Nvidia purchases Arm in $40 billion deal
"Therein lies the weakness," said Huang. "The ARM ecosystem is incredibly rich, but it's really rich for a specific device, an SOC," he continued.
"Now, you move into the world of open computing, and these types of environments require the rest of the platform; well, at Nvidia, we are really good at the rest of the platform, all the I/O, all the necessary system software."
ARM has had tremendous success with certain embedded offerings, such as its Mali GPU core. However, it has had limited success with selling neural network circuitry to embed into processors for AI processing. It has also had very little luck taking share from Intel x86 chips in data center servers.
Huang expects the broad technology stack Nvidia offers will improve upon that record.
Also: Nvidia unveils Jetson Nano 2GB for students and robot hobbyists
Huang compared ARM to x86, saying that Intel's processors dominate general purpose computing precisely because they are general and therefore have multiple uses. "You can download new drivers and it's a gaming PC, but it can also be an enterprise desktop, or you can add memory and it becomes a workstation, and then the next day it's a supercomputer."
The "apps we care about," such as artificial intelligence, need that broad, comprehensive computing platform, said Huang. "We have the stack, we'll make it happen," he said. "We will leverage the ecosystem and bring our soft IP to [ARM's] licensees.
Asked by a reporter why Nvidia needs to own ARM, rather than, say, partnering, Huang said Nvidia needs to control the distribution channel that ARM has built over 30 years to achieve its goals and also to get the financial benefit of it.
"In order to turbo-charge that success, we have to bring the Nvidia stack to the ARM [distribution channel], and I won't do that unless I can also capture the value that's created."
"So, I would buy the company first, and then make it more valuable."