NVIDIA dévoile les technologies d'IA et de GPU de nouvelle génération au CES 2025
NVIDIA, the powerhouse in graphics and AI technology, made a splash at CES 2025 with a series of groundbreaking announcements that promise to revolutionize computing, gaming, intelligence artificielle, and robotics. CEO Jensen Huang took center stage to unveil the company’s latest innovations, setting the stage for a new era of technological advancement.
Série RTX 50 : Redéfinir la technologie GPU
The spotlight of NVIDIA’s presentation was the introduction of the RTX 50 Series, powered by the new Blackwell architecture.
Cette prochaine génération GPU family brings unprecedented performance and AI capabilities to both desktop and laptop computers.
La série RTX 50 possède des fonctionnalités impressionnantes, notamment la RTX 5070 offrant des performances équivalentes à la 4090 à seulement 549 $, et la RTX 5090 délivrant deux fois la performance de l'ancien modèle 4090.
La série intègre des charges de travail d'IA avec le traitement graphique informatique, introduisant des innovations telles que la compression neurot de textures et l’ombrage neuromatériel.
NVIDIA a réussi à intégrer ces puissants GPU dans des ordinateurs portables, avec le modèle portable 5070 atteignant les performances d'une 4090 de bureau à partir de 1 299 $.
The future of computer graphics, according to Huang, is neuro rendering – the fusion of artificial intelligence and computer graphics.
Agents IA : L'avenir de l'informatique d'entreprise
A significant portion of Huang’s presentation focused on l'IA agentique for applications d'entreprise. NVIDIA introduced three key components to help build agentic AI:
- NVIDIA Nims: These are AI microservices packaged in containers, optimized for various applications such as vision, compréhension du langage, speech, animation, and digital biology. These models can run in any cloud environment, allowing easy integration into software packages to create AI agents for various enterprise applications.
- NVIDIA Nemo : Il s'agit d'un système d'intégration, de formation et d'évaluation d'employés numériques, conçu pour simplifier le processus d'intégration de l'IA dans les flux de travail d'entreprise.
- Llama Neotron Modèles linguistiques: NVIDIA announced a family of open-source language models based on Llama, fine-tuned for enterprise use. These models range from small, fast-response versions to ultra-large models capable of complex tasks. The Llama Neotron Suite includes models optimized for various enterprise applications, from quick interactions to complex problem-solving and even teacher models for training other AIs.
Huang a souligné que ces modèles d'IA are designed to be versatile and adaptable, capable of running in any cloud environment.
This flexibility allows for easy integration into existing software packages, enabling the creation of AI agents for a wide range of enterprise applications, from service client à l'analyse de données complexe.
NVIDIA Cosmos : Comprendre le monde physique
Peut-être l'annonce la plus ambitieuse était NVIDIA Cosmos, un modèle de fondation mondial conçu pour comprendre le monde physique.
Trained on 20 million hours of video, Cosmos focuses on physical dynamic things and has potential applications in robotics, génération de données synthétiques, and physical scenario prediction.
Cosmos features an auto-regressive model for real-time applications, a diffusion model for high-quality génération d'image, and a unique tokenizer that learns the vocabulary of the real world.
When connected to NVIDIA’s Omniverse platform, Cosmos can provide physically grounded, truth-conditioned AI generations, opening up new possibilities for simulation and prediction in various industries.
Véhicules autonomes : Conduire vers l'avenir
NVIDIA a réaffirmé sa position dans le véhicule autonome (AV) market with several key developments. The company announced partnerships with major automakers, including Toyota, for next-generation AVs.
Ils ont également présenté Thor, un nouveau ordinateur robotique pour véhicules avec une puissance de traitement 20 fois supérieure à celle de son prédécesseur.
Importantly, NVIDIA’s Drive OS has been certified to ASIL D, the highest standard of functional safety for automobiles, making it the first software-defined, programmable AI computer to achieve this certification.
Projet Digits : Supercalculateur IA en format compact
The keynote concluded with the unveiling of Project Digits, NVIDIA’s latest AI supercomputer.
Cet appareil compact, alimenté par la puce GB110, promet d'apporter des capacités de supercalcul à des ordinateurs de bureau.
Intégré dans collaboration with MediaTek, Project Digits runs the entire NVIDIA AI stack and is expected to be available around May 2025.
L'avenir de la robotique
Huang also touched on the future of robotics, suggesting that the “ChatGPT moment” for general la robotique est imminente.
NVIDIA’s Isaac platform aims to provide technology elements to the robotics industry to accelerate the development of general robotics.
NVIDIA’s CES 2025 presentation demonstrates the company’s continued leadership in GPU technology and AI innovation. With advancements spanning gaming, enterprise AI, autonomous vehicles, and supercomputing, NVIDIA is poised to shape the technological landscape for years to come.
The focus on AI agents and the introduction of Cosmos and Project Digits highlight NVIDIA’s commitment to pushing the boundaries of what’s possible in AI and computing, promising a future where intelligent systems are more integrated into our daily lives than ever before.