At CES 2025, Jensen Huang, the founder of NVIDIA, unveiled a series of spectacular product launches in his keynote speech. Among them, the debut of the new generation RTX 5090 graphics card undoubtedly attracted a great deal of attention, but NVIDIA’s ambitions go far beyond that.
RTX 50 Series: More Power, Less Price
Firstly, NVIDIA officially launched the next-generation RTX Blackwell GPU series, featuring four models: RTX 5090, RTX 5080, RTX 5070 Ti, and RTX 5070. These graphics cards come with a new Founders Edition design, equipped with dual-way circulation fans and GDDR7 memory, supporting PCIe Gen 5 and DisplayPort 2.1b interfaces, capable of driving display outputs up to 8K/165Hz.
The RTX 5090 and RTX 5080 will go on sale on January 30, while the RTX 5070 Ti and RTX 5070 are set to launch in February. Thanks to the new architecture Blackwell and DLSS 4 technology, the RTX 5090 is said to be twice as fast as its predecessor, the RTX 4090, with a corresponding increase in power consumption of 125 watts, but NVIDIA emphasizes its higher efficiency.
Omniverse + Cosmos: World Models and Digital Twins
Apart from graphics cards, NVIDIA also showcased its advancements in the field of world models. Huang introduced a series of world foundational models named Cosmos, which can predict and generate “physics-aware” videos, aiming to address the issue that current AI models can only “know the what, but not the why.”
The Cosmos models are divided into three categories: Nano, Super, and Ultra, with parameter scales ranging from 4 billion to 14 billion. These models are trained on massive real-world video data and can generate synthetic data for robotics, autonomous vehicles, and other fields.
NVIDIA also launched the Mega Omniverse Blueprint, a framework for creating industrial digital twins. This framework brings a new platform for industrial AI and robotics simulation to factories and warehouses through software-defined testing and optimization.
Project DIGITS: Supercomputing Center at Home
Addressing the demand for high computing power from individual developers and regular users, NVIDIA introduced a computing unit called Project DIGITS. This Mac mini-sized device can handle AI models with up to 200 billion parameters, powered by a standard household power outlet. It is equipped with NVIDIA’s next-generation GB10 Grace Blackwell chip, Grace CPU, 128GB of unified memory, and up to 4TB of NVMe storage space. Two Project DIGITS systems can be connected together to handle models with up to 405 billion parameters.
Project DIGITS is set to launch in May this year, with a starting price of $3,000, providing individual developers with a solution to access supercomputing power.
NVIDIA’s announcements not only showcase the company’s mature “AI infrastructure trilogy” in terms of computing power, models, and application landing, but also reveal its larger goals and ambitions. In the new wave of AI, the rapid evolution of computing power is undoubtedly the core, and NVIDIA is firmly holding the reins of this development. The products unveiled today are not just a testament to technological prowess, but also a promise of a future where the boundaries of what’s possible are constantly being expanded, blending the rigor of science with the warmth of human endeavor.
