Key Point 1: GeForce RTX 50 Series GPUs with Blackwell Architecture Unveiled
At CES, Jensen Huang introduced the GeForce RTX 50 series GPUs powered by the new Blackwell architecture, including the RTX 5090, RTX 5080, RTX 5070 Ti, and RTX 5070. Huang revealed that Blackwell is in full production, with GPUs set to hit the market later this January.
Huang showcased the RTX 5090 GPU, boasting 92 billion transistors and offering 3,400 TOPS of computational power (1 TOPS equals one trillion operations per second). The entry-level RTX 5070 delivers 1,000 TOPS of AI performance.
The new GPUs deliver unprecedented realism in gaming visuals. Pricing for the RTX 50 series was also announced: $1,999 for the RTX 5090, $999 for the RTX 5080, $749 for the RTX 5070 Ti, and $549 for the RTX 5070. Huang highlighted that the $549 RTX 5070 matches the performance of the $1,599 RTX 4090.
Huang also announced the development of the Grace Blackwell NVLink72 mega-chip, which integrates 72 Blackwell GPUs or 144 chips, surpassing the computational capabilities of the world’s fastest supercomputers. The NVLink system is designed for Agentic AI, enabling enhanced testing durations and improved customer interaction. Huang emphasized the unprecedented scale and computational strength of the Blackwell system, designed to push technological and innovative boundaries.
Key Point 2: Introducing the World Model Cosmos
Huang unveiled the Cosmos foundation model, capable of translating images and text into actionable tasks for robots. By seamlessly combining visual and linguistic comprehension, Cosmos executes complex actions. The model is open-source and available for download on GitHub.
A demo showed Cosmos responding to text, image, or video prompts, generating virtual world scenarios tailored to autonomous driving and robotics applications. For instance, instead of deploying fleets to collect real-world data or having humanoid robots trained through repeated human demonstrations, future training can rely on hyper-realistic virtual imagery generated by Cosmos.
However, Huang cautioned that Cosmos requires vast amounts of data to reach a "ChatGPT milestone" in AI capability.
Key Point 3: Launching the Thor Automotive Processor
Huang announced the next-generation automotive processor, Thor, a revolutionary robotics computer designed to process vast amounts of sensor data.
Key Point 4: Partnerships with Leading Automakers
Huang highlighted NVIDIA's automotive partnerships, including collaborations with Chinese automakers BYD, Li Auto, Xiaomi, and Zeekr. He announced that Toyota would adopt NVIDIA's automotive chips and operating systems and pledged to connect more developers and partners to Omniverse to expand the ecosystem. This aims to address significant inefficiencies and automation opportunities within the $50 trillion global GDP.
Key Point 5: Personal AI Supercomputer with MediaTek
Huang introduced Project DIGITS, a new high-performance desktop computer tailored for AI developers, accelerating AI system development and testing. This computer integrates NVIDIA's core data center chips with MediaTek's CPU technology and starts at $3,000, targeting professional markets.
Project DIGITS is ideal for testing AI models, especially in autonomous driving, robotics, and other physical AI applications, providing significant utility in these fields.