AI chips, robots, and reasoning models: 5 takeaways from Nvidia's GTC

Huang besides discussed the company’s enactment connected reasoning models, agentic AI, photonics, and robotics astatine the yearly developer conference, besides known arsenic the GTC.

Here are immoderate takeaways from Huang’s keynote.

Huang said Nvidia’s Blackwell chips — which helium announced astatine past year’s GTC — are present successful afloat production.

The adjacent Blackwell Ultra NVL72 chips, which person one-and-a-half times much representation and 2 times much bandwidth, volition beryllium utilized to accelerate gathering AI agents, carnal AI, and reasoning models, Huang said. Blackwell Ultra volition beryllium disposable successful the 2nd fractional of this year.

Nvidia’s next-generation GPUs named aft astronomer Vera Rubin volition beryllium released successful the 2nd fractional of adjacent year, Huang said. Rubin has a caller CPU that has doubly the show of its erstwhile chip, arsenic good arsenic much representation and bandwidth.

The 2nd fractional of 2027 volition bring Rubin Ultra, Huang said, which volition incorporate aggregate GPUs connected together.

Huang said Nvidia’s adjacent architecture volition beryllium named Feynman, and is coming successful 2028.

Nvidia Isaac GR00T N1 is the archetypal unfastened and afloat customizable AI instauration exemplary for gathering humanoid robots successful the world, Huang said.

The institution volition pre-train GR00T N1 connected synthetic and existent data, and merchandise the unfastened household of models to robotics developers.

Huang unveiled Nvidia Dynamo — “essentially the operating strategy of an AI factory” — to accelerate and standard reasoning models efficiently.

He besides announced the Llama Nemotron household of reasoning models which volition alteration developers and enterprises to physique AI agents. The caller reasoning models are open.

Huang introduced Nvidia DGX idiosyncratic AI supercomputers, which are powered by the company’s Blackwell AI platform. DGX Station and DGX Spark, formerly known arsenic Project DIGITS, volition alteration everyone from developers to students to prototype, fine-tune, and tally AI models from a idiosyncratic desktop.

For the latest news, Facebook, Twitter and Instagram.