Nvidia's latest AI chip packs more than 200 billion transistors
Published Date: 3/19/2024
Source: axios.com

Aiming to keep riding an AI-driven wave of demand, Nvidia is using its GTC developer conference this week to tout a more powerful family of chips along with a host of industry partnerships.

Why it matters: Nvidia has been the biggest beneficiary of the generative AI boom among chipmakers, but rivals are eager to get in on the action.


Driving the news: Nvidia unveiled its new Blackwell family of processors, due out later this year. Blackwell-architecture chips, which contain more than 200 billion transistors, are the successor to the company's current Hopper processors, which have been in short supply.

  • "Hopper is fantastic, but we need bigger GPUs," Nvidia CEO Jensen Huang said, per CNBC. Nvidia said Blackwell processors can power much larger language models than are deployed today — up to 27 trillion parameters, many times the size of today's most advanced models, like GPT-4.
  • Nvidia also used the event to highlight its expansion beyond chips into paid-for software, as well as its moves into new areas, including robots.
  • Other tech firms were eager to tout their ties to Nvidia, with companies including Google, Microsoft, Oracle, SAP and Hewlett Packard Enterprise announcing new or expanded partnerships.

The big picture: Nvidia, while the clear early leader in the chips that power generative AI, is facing pressure from established rivals, including AMD, as well as smaller upstart chipmakers focused specifically on handling artificial intelligence.

  • Even as they continue to purchase of Nvidia chips, tech giants including Microsoft, Google and Amazon are also developing their own processors capable of powering the AI systems within their own data centers.

What's next: Among the many sessions at GTC this week is a panel Wednesday bringing together the original authors of "Attention is All You Need," a seminal Google paper that introduced the transformer technology underlying most modern generative AI systems.

Go deeper: Meet the pickaxe vendors of the AI gold rush