Yesterday, NVIDIA CEO Jensen Huang addressed the company at the World Artificial Intelligence Conference. Naturally, the key note was focused on joining NVIDIA’s future AI technologies, including existing Blackwell processing architecture and beyond. Even Huang touched things like the use of AI in robotics, but NVIDIA stock prices are falling, the market does not affect much.
For the first part of the presentation, the RTX was hot on the Blackwell CEO’s lips. We have just seen the launch of NVIDIA’s new RTX 50 series cards that run new technology and DLSS 4 has been one of these few highlights. Between this and the AMD KFSR4, the use of AI scaling is the key to generations to help improve sports. Huang made a point to note the Server side version of Blackwell’s 40x increase in the AI ’factory’ performance on their hopper of NVIDIA.
So when the company did not announce any gaming PC, we showed two new Nvidia desktops. DGX Spark (former digits) and DGX station are desktop computer specifically designed to run AI. They can be used to run large models on hardware that are specifically designed for jobs. Possibly they will not choose choice for your next rig, such as enterprises also focuses on RTX Blackwell Profession announcedBut you can Enter interest for Golden AI BOCS.
Huang also covered the new NVIDIA’s new roadmap, which details the near -work work about AI. Since this is a developer aimed to conference, it is more about the team’s plan when you work in harmony with NVIDIA or is ready to use company technologies. It is being said that NVIDIA is making the next jump with Robin and Robin Ultra with extremely up -to -date capabilities.
Robin is the upcoming AI prepared architecture of NVIDIA, and it is not a sandwiches spell. Instead, this architecture is named after Vera Rubin, which astronomy discovered in the dark. It aims to introduce new design for CPU, and GPU as well as a memory system.
“Rubin is 900x Hopper’s performance is in the scale -up flop, which sets the next AI’s stage.” Huang says a crowd that is hopeful that it is understood.
It also comes with the Turbo Board version, Robin Ultra. It is for huge projects and will be able to create a rack up to 600 kW, which has more than two and a half million individual components per rack.
At both places in the market, NVIDIA hopes to face more demands. According to the Roadmap, we should see Rubin in the game for Nvidia in 2026, and then we will have a fanatic architecture, named after Physical Richard Fanman, in 2027-28
It will probably work together DinumoAnother Nvidia AI qualification was announced during the key. Dynamo, NVIDIA’s TriTon will be the successor of the Server Server, and is critically open source and available to all. The AI Inconing Services Software will help intervene between GPU and help language models, separate processing and reasoning works. It has already doubled the performance on Hopper, but it leads to a more interesting way to handle these tasks.
Dynamo stores what has been done, and will begin to allocate work to the GPU, which already contains information that can help. These GPUs will be more effective in these tasks. In fact, it seems a lot like how the real brain works. The more you think and combine them, the stronger the links will be and you will be better on ideas. But it is not just limited to GPUs and CPUs, but it is about to change storage.
“Instead of a recovered storage system, the future storage system is about to become a cement-based recovery system. A storage system that embedded raw data permanently in the background, and later, when you have access to it, you will not recover it and you will only ask it.” He explains.
If you are still not quarreling on it, get ready to be nvidia’s incredibly nominated Iszak Gr00t N1. Usually to become the first open and customized foundation model for humanitarian reasoning and expertise, it will teach your robot exactly what to do when an Apple falls on his head. Invent gravity.
“Nvidia Isaac Gr00t N1 and with new data generation and robot learning framework, robotics manufacturer will open the next border everywhere in the AI era.” Huang says.
It works by dividing the tasks into two different types, one for quick and sharp reactions, and the other is more thoughtful for argument. To do these works that look around the room and immediately analyze it, and then perform specific processes of specific robots. This is the first place in a series of modules that is planning to download and release NVIDIA.
These key notes are always very high for developers and enterprise users, rather than an average gamer, but they also point to future technologies that can wrap anywhere, including gaming. For Nvidia, it seems that we can expect the company to be involved in further AI development, and most of it seems to be being used well. Low AI for art purposes, more, more efficient storage, complex programming, and of course to improve graphics. How to catch a robot to teach a robot.
