Nvidia Annual Conference: new AI chips, PCs, robots

At the annual GTC conference, Nvidia announced new processors for creating and running AI models — Blackwell Ultra and Vera Rubin. Also presented was an engine for simulating robot movement and a number of other products.

The Blackwell Ultra chip family will be released in the second half of 2025. Vera Rubin is expected next year.

The architecture of the next generation processors ( after Rubin) will be named after physicist Richard Feynman. They will be available for sale in 2028.

Nvidia also introduced a number of other solutions, including:

• new desktop computers based on their own chips;
• Newton robot motion simulation engine and Groot N1 model;
• update of network components to integrate hundreds or thousands of graphics processors (GPU) into a single system;
• Dynamo software package that helps users get the most out of their chips.

Vera Rubin

The new system, named after the astronomer Vera Rubin, consists of two components: a central processor (CPU) called Vera and a GPU — Rubin.

Nvidia emphasized that it has designed the CPU based on the Olympus core for the first time on its own. Previously, the company used off-the-shelf Arm designs. Thanks to the new Vera solution, it is twice as fast compared to last year's Grace Blackwell chips.

In the pair with Vera, the Rubin graphics processor is capable of performing calculations at 50 petaflops (50 quadrillion operations per second). The current Blackwell has a rating of 20 petaflops. Rubin can also support up to 288 gigabytes of fast memory. This is one of the key parameters that artificial intelligence developers pay attention to.

Rubin technically consists of two GPUs in one case. Following it in 2027 will be Rubin Ultra. It will offer a set of four graphics processors, providing performance of up to 100 petaflops.

Blackwell Ultra

The new versions of flagship chips from the Blackwell family are capable of producing more tokens per second, allowing for the generation of a lot of content. As a result, providers will be able to offer premium AI services for time-sensitive applications.

Personal Computers

Nvidia has announced a new line of "personal supercomputers with artificial intelligence" based on the Grace Blackwell platform - DGX Spark (предыдущее called Project Digits) and DGX Station. They are capable of running large language models like Llama or DeepSeek.

"This is the computer of the AI era. This is how PCs should look. […] Now we have a whole range for enterprises, from small and tiny to workstations," noted the CEO of the American chip manufacturer Jensen Huang.

DGX Spark provides up to one quadrillion computations per second thanks to the 10 GB Grace Blackwell Superchip. The DGX Station is equipped with the GB300 Grace Blackwell combined with 784 GB of memory.

DGX Spark is available for order, and DGX Station will be released later in 2025 through various partners such as Asus, Boxx, Dell, HP, and Lenovo.

"AI agents will be everywhere. How they work, enterprises, and we — everything will change radically. Therefore, we need a new line of computers. And this is it," emphasized Huang.

Newton Robot Engine

Nvidia is collaborating with Disney Research and Google DeepMind on the development of Newton — a physics engine for simulating robot movements in real-world conditions.

Disney will be the first to apply a tool for its next-generation entertainment bots, such as BDX, inspired by "Star Wars." One of them walked on stage with Huang during the presentation.

Nvidia plans to release an early version of Newton with open source later in 2025.

The physics engine will help robots become more "expressive" and "learn to tackle complex tasks with greater accuracy," Nvidia said. Newton is designed to help developers simulate the interaction of bots with the natural world.

Model for Humanoids Groot N1

Separately, Nvidia announced the release of the Groot N1 model, which it calls "the foundation of AI for humanoid robotics." It features a "dual-system architecture" for "fast and slow thinking," inspired by human cognitive processes.

A slowly thinking system allows the robot to perceive and understand the surrounding environment and instructions, and then plan the correct actions. Fast thinking transforms the aforementioned plan into actions.

"The era of universal robotics has arrived," said Huang.

Groot N1 is the development of the Project Groot presented last year.

Partnership with General Motors

General Motors (GM) has entered into an expanded partnership with Nvidia to bring artificial intelligence into the physical world. The collaboration will impact all aspects of the automaker's business, including factories, robots, and self-driving cars.

The time for autonomous cars has come, emphasized Huang.

"We look forward to the opportunity to create artificial intelligence in all three areas together with GM. AI for manufacturing, so they can revolutionize the way production works; AI for enterprises, so they can revolutionize the way they design and model cars, as well as AI for machines," he said from the stage.

Nvidia will provide AI infrastructure and help GM create its own artificial intelligence. The automaker will also use Nvidia Drive AGX in its onboard equipment for future driver assistance systems and enhance driving safety.

Impact of DeepSeek

On January 20, the Chinese AI startup DeepSeek released a reasoning-oriented open-source model called DeepSeek-R1. The company offered performance comparable to top models at a much lower cost, raising doubts about the necessity of multi-billion dollar investments in the AI sector and leading to a collapse in Nvidia's stock.

Huang stated that DeepSeek has actually become a good signal for his company. The model of the Chinese firm uses a reasoning process that requires more computational power to provide better answers.

According to the CEO of Nvidia, Blackwell Ultra is well-suited for "thinking" models. The company has developed new chips for more efficient output processing.

Recall that Nvidia's financial report for the fourth quarter exceeded expectations, and the forecasts for the first quarter of 2025 turned out to be optimistic.