Nvdia's CES 2025 Event: Everything Revealed in 12 Minutes

Summary

Nvidia unveils groundbreaking advancements with its GeForce RTX 50 Series Blackwell architecture, delivering unprecedented GPU performance and AI capabilities. The new GPUs feature 92 billion transistors, 4,000 AI teraflops, and enhanced memory for next-level graphics and neural network processing. Additionally, Nvidia introduces the Thor processor for robotics and autonomous vehicles, and the Cosmos platform for AI training and multimodal learning. Project Digits, a cloud-based AI supercomputer, promises to accelerate AI research and development. Nvidia is paving the way for major breakthroughs in AI, robotics, and supercomputing, transforming industries and enabling the future of intelligent systems.

Takeaways

• 😀 NVIDIA introduces the GeForce RTX 50 Series Blackwell architecture, featuring 92 billion transistors and enhanced AI capabilities, including 4,000 TOPS and 4 petaflops of AI processing power.
• 😀 The RTX 50 Series provides groundbreaking performance improvements, offering 380 Ray Tracing teraflops for generating incredibly detailed images.
• 😀 The new architecture includes 125 Shader teraflops and introduces a dual-shader system—one for floating point and one for integer operations, enabling better performance in various workloads.
• 😀 Micron's G7 memory brings 1.8 terabytes per second of bandwidth, doubling the performance of the previous generation, significantly boosting AI and graphics workloads.
• 😀 RTX 50 GPUs can seamlessly mix AI workloads with computer graphics, and the programmable shaders can now process neural networks, leading to new breakthroughs like neurot texture compression and neural material shading.
• 😀 The RTX 570 and 590 models provide up to twice the performance of previous generations, with prices starting at $549, showing how NVIDIA has made high-performance GPUs more accessible.
• 😀 NVIDIA has successfully integrated RTX 5070-level performance into thin laptops, making 4090-level gaming and AI power portable with devices like the $1,299 5070 laptop.
• 😀 The Blackwell system achieves an astonishing 1.4 exaflops of AI floating-point performance, outperforming the world's fastest supercomputers and processing the equivalent of global internet traffic in data bandwidth.
• 😀 NVIDIA's new platform, Nvidia Cosmos, is an open-source, AI-powered model designed to understand the physical world and generate synthetic data, supporting advancements in robotics, language models, and autonomous vehicles.
• 😀 The Cosmos platform includes advanced features like an auto-regressive model for real-time applications, a diffusion model for high-quality image generation, and a data pipeline for efficient training with CUDA and AI acceleration.
• 😀 Thor, NVIDIA’s new AI processor for autonomous vehicles and robotics, delivers 20 times the processing power of its predecessor, Orin, and is in full production, marking a major leap forward in general robotics and AI systems.
• 😀 NVIDIA’s new project, a universal robotics computer dubbed 'Project Digits,' integrates AI software stacks into a supercomputer platform that can be accessed like a cloud service, with an upcoming release of the Grace Blackwell chip built in collaboration with Mediatek.

Q & A

What are the key improvements in the GeForce RTX 50 Series with Blackwell architecture?

The GeForce RTX 50 Series with Blackwell architecture offers significant advancements, including 92 billion transistors, 4,000 TOPS (Tera Operations Per Second), 4 PFLOPS of AI processing power, and 380 Ray Tracing teraflops. It also features 125 Shader teraflops and a dual-shader architecture, with floating point and integer shaders performing concurrently.

How does the memory bandwidth of the RTX 50 Series compare to previous generations?

The RTX 50 Series includes G7 memory from Micron, delivering 1.8 terabytes per second of bandwidth, which is twice the performance of the previous generation. This high bandwidth is crucial for handling AI workloads and complex graphics computations.

What is the significance of integrating AI workloads with computer graphics in the RTX 50 Series?

Integrating AI workloads with computer graphics allows for more efficient processing of complex tasks like rendering, image generation, and real-time AI applications. The RTX 50 Series can handle both graphics and AI tasks simultaneously, providing significant performance improvements for tasks that require neural network processing, such as neurot texture compression and neural material shading.

How does the RTX 50 Series support AI applications in its shader architecture?

The RTX 50 Series features a programmable shader capable of processing neural networks, enabling AI functionalities within the shader pipeline. This allows for innovations like neurot texture compression and neural material shading, enhancing the quality and performance of image generation.

What makes the Cosmos platform unique in AI development?

The Nvidia Cosmos platform is unique because it combines AI and CUDA acceleration in a data pipeline, enabling real-time applications and high-quality image generation. It uses a foundation model trained on 20 million hours of video to understand the physical world, which supports various downstream applications such as robotics, captioning, and synthetic data generation.

How does the Cosmos Foundation model contribute to AI's understanding of the physical world?

The Cosmos Foundation model is trained on video data focusing on dynamic physical themes such as humans walking, manipulating objects, and fast camera movements. This training helps the AI understand the physical world, which can be applied to robotics, image captioning, and training large language models, allowing for more realistic AI-generated content.

What is Nvidia Thor, and how does it differ from previous autonomous vehicle processors?

Nvidia Thor is the next-generation robotics processor for autonomous vehicles, boasting 20 times the processing capability of the previous generation, Orin. It processes massive amounts of sensor data, such as high-resolution cameras, radar, and LiDAR, to predict paths and enable real-time decision-making. Thor is also versatile and can be used in general robotics, not just autonomous vehicles.

What are the three types of robots mentioned as critical for the future of general robotics?

The three types of robots considered critical for the future of general robotics are: 1) Agentic robots, which function as information workers, 2) Self-driving cars, which benefit from the infrastructure of roads and cities, and 3) Human robots, which could operate in environments built for humans and have broader applications in various industries.

What is the significance of the 'Project Digits' AI supercomputer?

Project Digits is Nvidia's latest AI supercomputer that integrates the company's AI stack and runs on an advanced, secret chip known as GB 110. This supercomputer is designed for both cloud and workstation use and is expected to push the boundaries of AI processing, enabling significant breakthroughs in AI research and application.

How does Nvidia's collaboration with MediaTek impact the development of the GB 110 chip?

Nvidia's collaboration with MediaTek helped create the GB 110 chip, a small but powerful processor that combines a Grace CPU with a Blackwell GPU. This chip is designed for high-performance AI computing and will be central to Project Digits, enhancing the supercomputer's overall efficiency and capability.